llama-graph : fix recurrent state copy

The `state_copy` shuffle assumes everything is moved at once,
which is not true when `states_extra` is copied back to the cache
before copying the range of states between `head` and `head + n_seqs`.
This is only a problem if any of the cells in [`head`, `head + n_seqs`)
have an `src` in [`head + n_seqs`, `head + n_kv`),
which does happen when `n_ubatch > 1` in the `llama-parallel` example.

Changing the order of the operations avoids the potential overwrite
before use, although when copies are avoided (like with Mamba2),
this will require further changes.

* llama-graph : rename n_state to state_size in build_recurrent_state

This naming should reduce confusion between the state size
and the number of states.

.github/labeler.yml

@@ -86,3 +86,10 @@ nix:
 embedding:
     - changed-files:
         - any-glob-to-any-file: examples/embedding/
+
+Ascend NPU:
+    - changed-files:
+        - any-glob-to-any-file:
+            - ggml/include/ggml-cann.h
+            - ggml/src/ggml-cann/**
+            - docs/backend/CANN.md

.github/workflows/build-linux-cross.yml

@@ -231,3 +231,116 @@ jobs:
                          -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
 
           cmake --build build --config Release -j $(nproc)
+
+  debian-13-loongarch64-cpu-cross:
+    runs-on: ubuntu-24.04
+    container: debian@sha256:653dfb9f86c3782e8369d5f7d29bb8faba1f4bff9025db46e807fa4c22903671
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Setup LoongArch
+        run: |
+          rm -f /etc/apt/sources.list.d/*
+          cat << EOF | tee /etc/apt/sources.list.d/debian-ports.list
+          deb http://snapshot.debian.org/archive/debian/20250515T202920Z/ trixie main
+          EOF
+          ( echo 'quiet "true";'; \
+            echo 'APT::Get::Assume-Yes "true";'; \
+            echo 'APT::Install-Recommends "false";'; \
+            echo 'Acquire::Check-Valid-Until "false";'; \
+            echo 'Acquire::Retries "5";'; \
+          ) > /etc/apt/apt.conf.d/99snapshot-repos
+
+          apt-get update
+          apt-get install -y ca-certificates debian-ports-archive-keyring cmake git zip
+          dpkg --add-architecture loong64
+
+          # Add arch-specific repositories for non-amd64 architectures
+          cat << EOF | tee /etc/apt/sources.list.d/loong64-ports.list
+          deb [arch=loong64] http://snapshot.debian.org/archive/debian-ports/20250515T194251Z/ sid main
+          EOF
+
+          apt-get update || true    ;# Prevent failure due to missing URLs.
+
+          apt-get install -y --no-install-recommends \
+                  build-essential \
+                  gcc-14-loongarch64-linux-gnu \
+                  g++-14-loongarch64-linux-gnu
+
+      - name: Build
+        run: |
+          cmake -B build -DLLAMA_CURL=OFF \
+                         -DCMAKE_BUILD_TYPE=Release \
+                         -DGGML_OPENMP=OFF \
+                         -DLLAMA_BUILD_EXAMPLES=ON \
+                         -DLLAMA_BUILD_TOOLS=ON \
+                         -DLLAMA_BUILD_TESTS=OFF \
+                         -DCMAKE_SYSTEM_NAME=Linux \
+                         -DCMAKE_SYSTEM_PROCESSOR=loongarch64 \
+                         -DCMAKE_C_COMPILER=loongarch64-linux-gnu-gcc-14 \
+                         -DCMAKE_CXX_COMPILER=loongarch64-linux-gnu-g++-14 \
+                         -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+                         -DCMAKE_FIND_ROOT_PATH=/usr/lib/loongarch64-linux-gnu \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+
+          cmake --build build --config Release -j $(nproc)
+
+  debian-13-loongarch64-vulkan-cross:
+    runs-on: ubuntu-24.04
+    container: debian@sha256:653dfb9f86c3782e8369d5f7d29bb8faba1f4bff9025db46e807fa4c22903671
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Setup LoongArch
+        run: |
+          rm -f /etc/apt/sources.list.d/*
+          cat << EOF | tee /etc/apt/sources.list.d/debian-ports.list
+          deb http://snapshot.debian.org/archive/debian/20250515T202920Z/ trixie main
+          EOF
+          ( echo 'quiet "true";'; \
+            echo 'APT::Get::Assume-Yes "true";'; \
+            echo 'APT::Install-Recommends "false";'; \
+            echo 'Acquire::Check-Valid-Until "false";'; \
+            echo 'Acquire::Retries "5";'; \
+          ) > /etc/apt/apt.conf.d/99snapshot-repos
+
+          apt-get update
+          apt-get install -y ca-certificates debian-ports-archive-keyring cmake git zip
+          dpkg --add-architecture loong64
+
+          # Add arch-specific repositories for non-amd64 architectures
+          cat << EOF | tee /etc/apt/sources.list.d/loong64-ports.list
+          deb [arch=loong64] http://snapshot.debian.org/archive/debian-ports/20250515T194251Z/ sid main
+          EOF
+
+          apt-get update || true    ;# Prevent failure due to missing URLs.
+
+          apt-get install -y --no-install-recommends \
+                  build-essential \
+                  glslc \
+                  gcc-14-loongarch64-linux-gnu \
+                  g++-14-loongarch64-linux-gnu \
+                  libvulkan-dev:loong64
+
+      - name: Build
+        run: |
+          cmake -B build -DLLAMA_CURL=OFF \
+                         -DCMAKE_BUILD_TYPE=Release \
+                         -DGGML_VULKAN=ON \
+                         -DGGML_OPENMP=OFF \
+                         -DLLAMA_BUILD_EXAMPLES=ON \
+                         -DLLAMA_BUILD_TOOLS=ON \
+                         -DLLAMA_BUILD_TESTS=OFF \
+                         -DCMAKE_SYSTEM_NAME=Linux \
+                         -DCMAKE_SYSTEM_PROCESSOR=loongarch64 \
+                         -DCMAKE_C_COMPILER=loongarch64-linux-gnu-gcc-14 \
+                         -DCMAKE_CXX_COMPILER=loongarch64-linux-gnu-g++-14 \
+                         -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+                         -DCMAKE_FIND_ROOT_PATH=/usr/lib/loongarch64-linux-gnu \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+
+          cmake --build build --config Release -j $(nproc)

.github/workflows/build.yml

@@ -839,12 +839,12 @@ jobs:
               -DGGML_CUDA=ON
             cmake --build build
 
-  windows-2019-cmake-cuda:
-    runs-on: windows-2019
+  windows-2022-cmake-cuda:
+    runs-on: windows-2022
 
     strategy:
       matrix:
-        cuda: ['12.4', '11.7']
+        cuda: ['12.4']
 
     steps:
       - name: Clone
@@ -878,7 +878,7 @@ jobs:
         env:
           CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
         run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
           cmake -S . -B build -G "Ninja Multi-Config" ^
             -DLLAMA_BUILD_SERVER=ON ^
             -DGGML_NATIVE=OFF ^

.github/workflows/release.yml

@@ -131,8 +131,9 @@ jobs:
         include:
           - build: 'x64'
             os: ubuntu-22.04
-          - build: 'arm64'
-            os: ubuntu-22.04-arm
+          # GGML_BACKEND_DL and GGML_CPU_ALL_VARIANTS are not currently supported on arm
+          # - build: 'arm64'
+          #   os: ubuntu-22.04-arm
 
     runs-on: ${{ matrix.os }}
 
@@ -159,6 +160,9 @@ jobs:
         id: cmake_build
         run: |
           cmake -B build \
+            -DGGML_BACKEND_DL=ON \
+            -DGGML_NATIVE=OFF \
+            -DGGML_CPU_ALL_VARIANTS=ON \
             -DLLAMA_FATAL_WARNINGS=ON \
             ${{ env.CMAKE_ARGS }}
           cmake --build build --config Release -j $(nproc)
@@ -207,6 +211,9 @@ jobs:
         id: cmake_build
         run: |
           cmake -B build \
+            -DGGML_BACKEND_DL=ON \
+            -DGGML_NATIVE=OFF \
+            -DGGML_CPU_ALL_VARIANTS=ON \
             -DGGML_VULKAN=ON \
             ${{ env.CMAKE_ARGS }}
           cmake --build build --config Release -j $(nproc)
@@ -373,11 +380,11 @@ jobs:
           name: llama-bin-win-${{ matrix.backend }}-${{ matrix.arch }}.zip
 
   windows-cuda:
-    runs-on: windows-2019
+    runs-on: windows-2022
 
     strategy:
       matrix:
-        cuda: ['12.4', '11.7']
+        cuda: ['12.4']
 
     steps:
       - name: Clone
@@ -405,7 +412,7 @@ jobs:
         id: cmake_build
         shell: cmd
         run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
           cmake -S . -B build -G "Ninja Multi-Config" ^
             -DGGML_BACKEND_DL=ON ^
             -DGGML_NATIVE=OFF ^

.github/workflows/server.yml

@@ -180,7 +180,7 @@ jobs:
 
 
   server-windows:
-    runs-on: windows-2019
+    runs-on: windows-2022
 
     steps:
       - name: Clone

CMakeLists.txt

@@ -159,6 +159,11 @@ if (NOT TARGET ggml AND NOT LLAMA_USE_SYSTEM_GGML)
     # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()
 
+if (MINGW)
+    # Target Windows 8 for PrefetchVirtualMemory
+    add_compile_definitions(_WIN32_WINNT=${GGML_WIN_VER})
+endif()
+
 #
 # build the library
 #

Makefile

@@ -367,7 +367,7 @@ ifdef LLAMA_SERVER_SSL
 endif
 
 ifndef GGML_NO_CPU_AARCH64
-	MK_CPPFLAGS += -DGGML_USE_CPU_AARCH64
+	MK_CPPFLAGS += -DGGML_USE_CPU_REPACK
 endif
 
 # warnings
@@ -970,7 +970,7 @@ OBJ_GGML = \
 	$(DIR_GGML)/src/ggml-threading.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu_cpp.o \
-	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-aarch64.o \
+	$(DIR_GGML)/src/ggml-cpu/repack.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-hbm.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-quants.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-traits.o \

README.md

@@ -3,6 +3,7 @@
 ![llama](https://user-images.githubusercontent.com/1991296/230134379-7181e485-c521-4d23-a0d6-f7b3b61ba524.png)
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
+[![Release](https://img.shields.io/github/v/release/ggml-org/llama.cpp)](https://github.com/ggml-org/llama.cpp/releases)
 [![Server](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml)
 
 [Roadmap](https://github.com/users/ggerganov/projects/7) / [Project status](https://github.com/ggml-org/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggml-org/llama.cpp/discussions/205) / [ggml](https://github.com/ggml-org/ggml)
@@ -28,6 +29,30 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)
 
 ----
 
+## Quick start
+
+Getting started with llama.cpp is straightforward. Here are several ways to install it on your machine:
+
+- Install `llama.cpp` using [brew, nix or winget](docs/install.md)
+- Run with Docker - see our [Docker documentation](docs/docker.md)
+- Download pre-built binaries from the [releases page](https://github.com/ggml-org/llama.cpp/releases)
+- Build from source by cloning this repository - check out [our build guide](docs/build.md)
+
+Once installed, you'll need a model to work with. Head to the [Obtaining and quantizing models](#obtaining-and-quantizing-models) section to learn more.
+
+Example command:
+
+```sh
+# Use a local model file
+llama-cli -m my_model.gguf
+
+# Or download and run a model directly from Hugging Face
+llama-cli -hf ggml-org/gemma-3-1b-it-GGUF
+
+# Launch OpenAI-compatible API server
+llama-server -hf ggml-org/gemma-3-1b-it-GGUF
+```
+
 ## Description
 
 The main goal of `llama.cpp` is to enable LLM inference with minimal setup and state-of-the-art performance on a wide
@@ -230,6 +255,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 
 </details>
 
+
 ## Supported backends
 
 | Backend | Target devices |
@@ -246,16 +272,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 | [OpenCL](docs/backend/OPENCL.md) | Adreno GPU |
 | [RPC](https://github.com/ggml-org/llama.cpp/tree/master/tools/rpc) | All |
 
-## Building the project
-
-The main product of this project is the `llama` library. Its C-style interface can be found in [include/llama.h](include/llama.h).
-The project also includes many example programs and tools using the `llama` library. The examples range from simple, minimal code snippets to sophisticated sub-projects such as an OpenAI-compatible HTTP server. Possible methods for obtaining the binaries:
-
-- Clone this repository and build locally, see [how to build](docs/build.md)
-- On MacOS or Linux, install `llama.cpp` via [brew, flox or nix](docs/install.md)
-- Use a Docker image, see [documentation for Docker](docs/docker.md)
-- Download pre-built binaries from [releases](https://github.com/ggml-org/llama.cpp/releases)
-
 ## Obtaining and quantizing models
 
 The [Hugging Face](https://huggingface.co) platform hosts a [number of LLMs](https://huggingface.co/models?library=gguf&sort=trending) compatible with `llama.cpp`:
@@ -263,7 +279,11 @@ The [Hugging Face](https://huggingface.co) platform hosts a [number of LLMs](htt
 - [Trending](https://huggingface.co/models?library=gguf&sort=trending)
 - [LLaMA](https://huggingface.co/models?sort=trending&search=llama+gguf)
 
-You can either manually download the GGUF file or directly use any `llama.cpp`-compatible models from [Hugging Face](https://huggingface.co/) or other model hosting sites, such as [ModelScope](https://modelscope.cn/), by using this CLI argument: `-hf <user>/<model>[:quant]`.
+You can either manually download the GGUF file or directly use any `llama.cpp`-compatible models from [Hugging Face](https://huggingface.co/) or other model hosting sites, such as [ModelScope](https://modelscope.cn/), by using this CLI argument: `-hf <user>/<model>[:quant]`. For example:
+
+```sh
+llama-cli -hf ggml-org/gemma-3-1b-it-GGUF
+```
 
 By default, the CLI would download from Hugging Face, you can switch to other options with the environment variable `MODEL_ENDPOINT`. For example, you may opt to downloading model checkpoints from ModelScope or other model sharing communities by setting the environment variable, e.g. `MODEL_ENDPOINT=https://www.modelscope.cn/`.
 

ci/run.sh

@@ -46,7 +46,20 @@ if [ ! -z ${GG_BUILD_METAL} ]; then
 fi
 
 if [ ! -z ${GG_BUILD_CUDA} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_CUDA=ON"
+
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        CUDA_ARCH=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader,nounits 2>/dev/null | head -1 | tr -d '.')
+        if [[ -n "$CUDA_ARCH" && "$CUDA_ARCH" =~ ^[0-9]+$ ]]; then
+            CMAKE_EXTRA="${CMAKE_EXTRA} -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCH}"
+        else
+            echo "Warning: Using fallback CUDA architectures"
+            CMAKE_EXTRA="${CMAKE_EXTRA} -DCMAKE_CUDA_ARCHITECTURES=61;70;75;80;86;89"
+        fi
+    else
+        echo "Error: nvidia-smi not found, cannot build with CUDA"
+        exit 1
+    fi
 fi
 
 if [ ! -z ${GG_BUILD_SYCL} ]; then

common/common.cpp

@@ -934,7 +934,7 @@ struct common_init_result common_init_from_params(common_params & params) {
         return iparams;
     }
 
-    if (params.ctx_shift && !llama_kv_self_can_shift(lctx)) {
+    if (params.ctx_shift && !llama_memory_can_shift(llama_get_memory(lctx))) {
         LOG_WRN("%s: KV cache shifting is not supported for this context, disabling KV cache shifting\n", __func__);
         params.ctx_shift = false;
     }
@@ -1041,7 +1041,7 @@ struct common_init_result common_init_from_params(common_params & params) {
         if (llama_model_has_decoder(model)) {
             llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch)));
         }
-        llama_kv_self_clear(lctx);
+        llama_memory_clear(llama_get_memory(lctx), true);
         llama_synchronize(lctx);
         llama_perf_context_reset(lctx);
         llama_set_warmup(lctx, false);

common/speculative.cpp

@@ -144,6 +144,8 @@ llama_tokens common_speculative_gen_draft(
     auto & smpl   = spec->smpl;
     auto & prompt = spec->prompt;
 
+    auto * mem = llama_get_memory(ctx);
+
     int reuse_i = 0;
     int reuse_n = 0;
 
@@ -173,7 +175,7 @@ llama_tokens common_speculative_gen_draft(
     result.reserve(params.n_draft);
 
     if (reuse_n == 0) {
-        llama_kv_self_clear(ctx);
+        llama_memory_clear(mem, false);
 
         prompt.clear();
     } else {
@@ -192,14 +194,14 @@ llama_tokens common_speculative_gen_draft(
         }
 
         if (reuse_i > 0) {
-            llama_kv_self_seq_rm (ctx, 0, 0, reuse_i);
-            llama_kv_self_seq_add(ctx, 0, reuse_i, -1, -reuse_i);
+            llama_memory_seq_rm (mem, 0, 0, reuse_i);
+            llama_memory_seq_add(mem, 0, reuse_i, -1, -reuse_i);
 
             prompt.erase(prompt.begin(), prompt.begin() + reuse_i);
         }
 
         if (reuse_n < (int) prompt.size()) {
-            llama_kv_self_seq_rm (ctx, 0, reuse_n, -1);
+            llama_memory_seq_rm (mem, 0, reuse_n, -1);
 
             prompt.erase(prompt.begin() + reuse_n, prompt.end());
         }

convert_hf_to_gguf.py

@@ -3709,8 +3709,7 @@ class BertModel(TextModel):
         self._try_set_pooling_type()
 
         if self.cls_out_labels:
-            key_name = gguf.Keys.Classifier.OUTPUT_LABELS.format(arch = gguf.MODEL_ARCH_NAMES[self.model_arch])
-            self.gguf_writer.add_array(key_name, [v for k, v in sorted(self.cls_out_labels.items())])
+            self.gguf_writer.add_classifier_output_labels([v for k, v in sorted(self.cls_out_labels.items())])
 
     def set_vocab(self):
         tokens, toktypes, tokpre = self.get_vocab_base()

docs/backend/CANN.md

@@ -8,6 +8,7 @@
  - [DataType Supports](#datatype-supports)
  - [Docker](#docker)
  - [Linux](#linux)
+ - [Environment variable setup](#environment-variable-setup)
  - [TODO](#todo)
 
 
@@ -290,5 +291,24 @@ Authors from Peking University: Bizhao Shi (bshi@pku.edu.cn), Yuxin Yang (yxyang
 
 We would like to thank Tuo Dai, Shanni Li, and all of the project maintainers from Huawei Technologies Co., Ltd for their help during the code development and pull request.
 
+## Environment variable setup
+
+### GGML_CANN_ASYNC_MODE
+
+Enables asynchronous operator submission. Disabled by default.
+
+### GGML_CANN_MEM_POOL
+
+Specifies the memory pool management strategy:
+
+- vmm: Utilizes a virtual memory manager pool. If hardware support for VMM is unavailable, falls back to the legacy (leg) memory pool.
+
+- prio: Employs a priority queue-based memory pool management.
+- leg: Uses a fixed-size buffer pool.
+
+### GGML_CANN_DISABLE_BUF_POOL_CLEAN
+
+Controls automatic cleanup of the memory pool. This option is only effective when using the prio or leg memory pool strategies.
+
 ## TODO
 - Support more models and data types.

docs/build.md

@@ -1,5 +1,9 @@
 # Build llama.cpp locally
 
+The main product of this project is the `llama` library. Its C-style interface can be found in [include/llama.h](include/llama.h).
+
+The project also includes many example programs and tools using the `llama` library. The examples range from simple, minimal code snippets to sophisticated sub-projects such as an OpenAI-compatible HTTP server.
+
 **To get the Code:**
 
 ```bash

docs/install.md

@@ -1,28 +1,42 @@
 # Install pre-built version of llama.cpp
 
-## Homebrew
+| Install via | Windows | Mac | Linux |
+|-------------|---------|-----|-------|
+| Winget      | ✅      |      |      |
+| Homebrew    |         | ✅   | ✅   |
+| MacPorts    |         | ✅   |      |
+| Nix         |         | ✅   | ✅   |
 
-On Mac and Linux, the homebrew package manager can be used via
+## Winget (Windows)
+
+```sh
+winget install llama.cpp
+```
+
+The package is automatically updated with new `llama.cpp` releases. More info: https://github.com/ggml-org/llama.cpp/issues/8188
+
+## Homebrew (Mac and Linux)
 
 ```sh
 brew install llama.cpp
 ```
+
 The formula is automatically updated with new `llama.cpp` releases. More info: https://github.com/ggml-org/llama.cpp/discussions/7668
 
-## MacPorts
+## MacPorts (Mac)
 
 ```sh
 sudo port install llama.cpp
 ```
-see also: https://ports.macports.org/port/llama.cpp/details/
 
-## Nix
+See also: https://ports.macports.org/port/llama.cpp/details/
 
-On Mac and Linux, the Nix package manager can be used via
+## Nix (Mac and Linux)
 
 ```sh
 nix profile install nixpkgs#llama-cpp
 ```
+
 For flake enabled installs.
 
 Or
@@ -34,13 +48,3 @@ nix-env --file '<nixpkgs>' --install --attr llama-cpp
 For non-flake enabled installs.
 
 This expression is automatically updated within the [nixpkgs repo](https://github.com/NixOS/nixpkgs/blob/nixos-24.05/pkgs/by-name/ll/llama-cpp/package.nix#L164).
-
-## Flox
-
-On Mac and Linux, Flox can be used to install llama.cpp within a Flox environment via
-
-```sh
-flox install llama-cpp
-```
-
-Flox follows the nixpkgs build of llama.cpp.

examples/batched.swift/Sources/main.swift

@@ -116,7 +116,7 @@ if llama_decode(context, batch) != 0 {
 }
 
 for i in 1 ..< n_parallel {
-    llama_kv_self_seq_cp(context, 0, Int32(i), 0, batch.n_tokens)
+    llama_memory_seq_cp(llama_get_memory(context), 0, Int32(i), 0, batch.n_tokens)
 }
 
 if n_parallel > 1 {

examples/embedding/embedding.cpp

@@ -37,7 +37,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
     const enum llama_pooling_type pooling_type = llama_pooling_type(ctx);
 
     // clear previous kv_cache values (irrelevant for embeddings)
-    llama_kv_self_clear(ctx);
+    llama_memory_clear(llama_get_memory(ctx), true);
 
     // run model
     LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);
@@ -236,9 +236,24 @@ int main(int argc, char ** argv) {
                 LOG("\n");
             }
         } else if (pooling_type == LLAMA_POOLING_TYPE_RANK) {
+            const uint32_t n_cls_out = llama_model_n_cls_out(model);
+            std::vector<std::string> cls_out_labels;
+
+            for (uint32_t i = 0; i < n_cls_out; i++) {
+                const char * label = llama_model_cls_label(model, i);
+                const std::string label_i(label == nullptr ? "" : label);
+                cls_out_labels.emplace_back(label_i.empty() ? std::to_string(i) : label_i);
+            }
+
             for (int j = 0; j < n_embd_count; j++) {
-                // NOTE: if you change this log - update the tests in ci/run.sh
-                LOG("rerank score %d: %8.3f\n", j, emb[j * n_embd]);
+                for (uint32_t i = 0; i < n_cls_out; i++) {
+                    // NOTE: if you change this log - update the tests in ci/run.sh
+                    if (n_cls_out == 1) {
+                        LOG("rerank score %d: %8.3f\n", j, emb[j * n_embd]);
+                    } else {
+                        LOG("rerank score %d: %8.3f [%s]\n", j, emb[j * n_embd + i], cls_out_labels[i].c_str());
+                    }
+                }
             }
         } else {
             // print the first part of the embeddings or for a single prompt, the full embedding

examples/gritlm/gritlm.cpp

@@ -45,7 +45,7 @@ static std::vector<std::vector<float>> encode(llama_context * ctx, const std::ve
         }
 
         // clear previous kv_cache values (irrelevant for embeddings)
-        llama_kv_self_clear(ctx);
+        llama_memory_clear(llama_get_memory(ctx), true);
         llama_set_embeddings(ctx, true);
         llama_set_causal_attn(ctx, false);
 
@@ -102,7 +102,7 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std
 
     llama_token eos_token = llama_vocab_eos(vocab);
 
-    llama_kv_self_clear(ctx);
+    llama_memory_clear(llama_get_memory(ctx), true);
     llama_set_embeddings(ctx, false);
     llama_set_causal_attn(ctx, true);
 

examples/llama.android/llama/src/main/cpp/llama-android.cpp

@@ -194,7 +194,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
         }
 
         batch->logits[batch->n_tokens - 1] = true;
-        llama_kv_self_clear(context);
+        llama_memory_clear(llama_get_memory(context), false);
 
         const auto t_pp_start = ggml_time_us();
         if (llama_decode(context, *batch) != 0) {
@@ -206,7 +206,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
 
         LOGi("Benchmark text generation (tg)");
 
-        llama_kv_self_clear(context);
+        llama_memory_clear(llama_get_memory(context), false);
         const auto t_tg_start = ggml_time_us();
         for (i = 0; i < tg; i++) {
 
@@ -223,7 +223,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
 
         const auto t_tg_end = ggml_time_us();
 
-        llama_kv_self_clear(context);
+        llama_memory_clear(llama_get_memory(context), false);
 
         const auto t_pp = double(t_pp_end - t_pp_start) / 1000000.0;
         const auto t_tg = double(t_tg_end - t_tg_start) / 1000000.0;
@@ -448,5 +448,5 @@ Java_android_llama_cpp_LLamaAndroid_completion_1loop(
 extern "C"
 JNIEXPORT void JNICALL
 Java_android_llama_cpp_LLamaAndroid_kv_1cache_1clear(JNIEnv *, jobject, jlong context) {
-    llama_kv_self_clear(reinterpret_cast<llama_context *>(context));
+    llama_memory_clear(llama_get_memory(reinterpret_cast<llama_context *>(context)), true);
 }

examples/llama.swiftui/llama.cpp.swift/LibLlama.swift

@@ -210,7 +210,7 @@ actor LlamaContext {
             }
             batch.logits[Int(batch.n_tokens) - 1] = 1 // true
 
-            llama_kv_self_clear(context)
+            llama_memory_clear(llama_get_memory(context), false)
 
             let t_pp_start = DispatchTime.now().uptimeNanoseconds / 1000;
 
@@ -223,7 +223,7 @@ actor LlamaContext {
 
             // bench text generation
 
-            llama_kv_self_clear(context)
+            llama_memory_clear(llama_get_memory(context), false)
 
             let t_tg_start = DispatchTime.now().uptimeNanoseconds / 1000;
 
@@ -242,7 +242,7 @@ actor LlamaContext {
 
             let t_tg_end = DispatchTime.now().uptimeNanoseconds / 1000;
 
-            llama_kv_self_clear(context)
+            llama_memory_clear(llama_get_memory(context), false)
 
             let t_pp = Double(t_pp_end - t_pp_start) / 1000000.0
             let t_tg = Double(t_tg_end - t_tg_start) / 1000000.0
@@ -292,7 +292,7 @@ actor LlamaContext {
     func clear() {
         tokens_list.removeAll()
         temporary_invalid_cchars.removeAll()
-        llama_kv_self_clear(context)
+        llama_memory_clear(llama_get_memory(context), true)
     }
 
     private func tokenize(text: String, add_bos: Bool) -> [llama_token] {

examples/lookahead/lookahead.cpp

@@ -60,6 +60,8 @@ int main(int argc, char ** argv) {
     llama_model * model = llama_init.model.get();
     llama_context * ctx = llama_init.context.get();
 
+    auto * mem = llama_get_memory(ctx);
+
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
     // Tokenize the prompt
@@ -94,7 +96,7 @@ int main(int argc, char ** argv) {
     llama_decode(ctx, llama_batch_get_one(&inp.back(),           1));
 
     for (int s = 1; s < W + G + 1; ++s) {
-        llama_kv_self_seq_cp(ctx, 0, s, -1, -1);
+        llama_memory_seq_cp(mem, 0, s, -1, -1);
     }
 
     const auto t_enc_end = ggml_time_us();
@@ -427,17 +429,17 @@ int main(int argc, char ** argv) {
 
         // KV cache management
         // if no verification token matched, we simply remove all cells from this batch -> no fragmentation
-        llama_kv_self_seq_rm(ctx, -1, n_past, -1);
+        llama_memory_seq_rm(mem, -1, n_past, -1);
 
         if (seq_id_best != 0) {
             // if a verification token matched, we keep the best sequence and remove the rest
             // this leads to some KV cache fragmentation
-            llama_kv_self_seq_keep(ctx, seq_id_best);
-            llama_kv_self_seq_cp  (ctx, seq_id_best, 0, -1, -1);
-            llama_kv_self_seq_rm  (ctx, seq_id_best,    -1, -1);
+            llama_memory_seq_keep(mem, seq_id_best);
+            llama_memory_seq_cp  (mem, seq_id_best, 0, -1, -1);
+            llama_memory_seq_rm  (mem, seq_id_best,    -1, -1);
 
             for (int s = 1; s < W + G + 1; ++s) {
-                llama_kv_self_seq_cp(ctx, 0, s, -1, -1);
+                llama_memory_seq_cp(mem, 0, s, -1, -1);
             }
         }
     }

examples/lookup/lookup.cpp

@@ -181,7 +181,7 @@ int main(int argc, char ** argv){
 
         // KV cache management
         // clean the cache of draft tokens that weren't accepted
-        llama_kv_self_seq_rm(ctx, 0, n_past, -1);
+        llama_memory_seq_rm(llama_get_memory(ctx), 0, n_past, -1);
 
         common_batch_clear(batch_tgt);
         common_batch_add(batch_tgt, draft[0], n_past, { 0 }, true);

examples/parallel/parallel.cpp

@@ -194,6 +194,8 @@ int main(int argc, char ** argv) {
     llama_model * model = llama_init.model.get();
     llama_context * ctx = llama_init.context.get();
 
+    auto * mem = llama_get_memory(ctx);
+
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
     // load the prompts from an external file if there are any
@@ -259,7 +261,7 @@ int main(int argc, char ** argv) {
 
         // assign the system KV cache to all parallel sequences
         for (int32_t i = 1; i <= n_clients; ++i) {
-            llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
+            llama_memory_seq_cp(mem, 0, i, -1, -1);
         }
 
         LOG_INF("\n");
@@ -286,9 +288,9 @@ int main(int argc, char ** argv) {
         if (batch.n_tokens == 0) {
             // all sequences have ended - clear the entire KV cache
             for (int i = 1; i <= n_clients; ++i) {
-                llama_kv_self_seq_rm(ctx, i, -1, -1);
+                llama_memory_seq_rm(mem, i, -1, -1);
                 // but keep the system prompt
-                llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
+                llama_memory_seq_cp(mem, 0, i, -1, -1);
             }
 
             LOG_INF("%s: clearing the KV cache\n", __func__);
@@ -447,8 +449,8 @@ int main(int argc, char ** argv) {
                     }
 
                     // delete only the generated part of the sequence, i.e. keep the system prompt in the cache
-                    llama_kv_self_seq_rm(ctx,    client.id + 1, -1, -1);
-                    llama_kv_self_seq_cp(ctx, 0, client.id + 1, -1, -1);
+                    llama_memory_seq_rm(mem,    client.id + 1, -1, -1);
+                    llama_memory_seq_cp(mem, 0, client.id + 1, -1, -1);
 
                     const auto t_main_end = ggml_time_us();
 

examples/passkey/passkey.cpp

@@ -126,6 +126,8 @@ int main(int argc, char ** argv) {
 
     int n_past = 0;
 
+    auto * mem = llama_get_memory(ctx);
+
     // fill the KV cache
     for (int i = 0; i < n_ctx; i += n_batch) {
         if (i > 0 && n_grp > 1) {
@@ -133,10 +135,10 @@ int main(int argc, char ** argv) {
             const int ib = i/n_batch - 1;
             const int bd = n_batch_grp*(n_grp - 1);
 
-            llama_kv_self_seq_add(ctx, 0, n_past - n_batch,         n_past,         ib*bd);
-            llama_kv_self_seq_div(ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
+            llama_memory_seq_add(mem, 0, n_past - n_batch,         n_past,         ib*bd);
+            llama_memory_seq_div(mem, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
 
-            n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
+            n_past = llama_memory_seq_pos_max(mem, 0) + 1;
         }
 
         common_batch_clear(batch);
@@ -166,10 +168,10 @@ int main(int argc, char ** argv) {
 
         LOG_INF("%s: shifting KV cache with %d\n", __func__, n_discard);
 
-        llama_kv_self_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
-        llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
+        llama_memory_seq_rm (mem, 0, n_keep            , n_keep + n_discard);
+        llama_memory_seq_add(mem, 0, n_keep + n_discard, n_ctx,  -n_discard);
 
-        n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
+        n_past = llama_memory_seq_pos_max(mem, 0) + 1;
 
         common_batch_clear(batch);
 
@@ -195,10 +197,10 @@ int main(int argc, char ** argv) {
         if (n_discard > 0) {
             LOG_INF("%s: shifting KV cache with %d to free space for the answer\n", __func__, n_discard);
 
-            llama_kv_self_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
-            llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
+            llama_memory_seq_rm (mem, 0, n_keep            , n_keep + n_discard);
+            llama_memory_seq_add(mem, 0, n_keep + n_discard, n_ctx,  -n_discard);
 
-            n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
+            n_past = llama_memory_seq_pos_max(mem, 0) + 1;
         }
     }
 

examples/retrieval/retrieval.cpp

@@ -83,7 +83,7 @@ static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & toke
 
 static void batch_process(llama_context * ctx, llama_batch & batch, float * output, int n_seq, int n_embd) {
     // clear previous kv_cache values (irrelevant for embeddings)
-    llama_kv_self_clear(ctx);
+    llama_memory_clear(llama_get_memory(ctx), false);
 
     // run model
     LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);

examples/save-load-state/save-load-state.cpp

@@ -196,7 +196,7 @@ int main(int argc, char ** argv) {
         fprintf(stderr, "%s : seq 0 copied, %zd bytes\n", __func__, ncopy);
 
         // erase whole kv
-        llama_kv_self_clear(ctx3);
+        llama_memory_clear(llama_get_memory(ctx3), true);
         fprintf(stderr, "%s : kv cache cleared\n", __func__);
 
         // restore kv into seq 1

examples/simple-chat/simple-chat.cpp

@@ -98,7 +98,7 @@ int main(int argc, char ** argv) {
     auto generate = [&](const std::string & prompt) {
         std::string response;
 
-        const bool is_first = llama_kv_self_seq_pos_max(ctx, 0) == 0;
+        const bool is_first = llama_memory_seq_pos_max(llama_get_memory(ctx), 0) == 0;
 
         // tokenize the prompt
         const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true);
@@ -113,7 +113,7 @@ int main(int argc, char ** argv) {
         while (true) {
             // check if we have enough space in the context to evaluate this batch
             int n_ctx = llama_n_ctx(ctx);
-            int n_ctx_used = llama_kv_self_seq_pos_max(ctx, 0);
+            int n_ctx_used = llama_memory_seq_pos_max(llama_get_memory(ctx), 0);
             if (n_ctx_used + batch.n_tokens > n_ctx) {
                 printf("\033[0m\n");
                 fprintf(stderr, "context size exceeded\n");

examples/speculative-simple/speculative-simple.cpp

@@ -217,7 +217,7 @@ int main(int argc, char ** argv) {
         {
             LOG_DBG("clear kv cache from any extra tokens, n_past = %d\n", n_past);
 
-            llama_kv_self_seq_rm(ctx_tgt, 0, n_past, -1);
+            llama_memory_seq_rm(llama_get_memory(ctx_tgt), 0, n_past, -1);
         }
 
         if ((params.n_predict >= 0 && n_predict > params.n_predict) || has_eos) {

examples/speculative/speculative.cpp

@@ -142,6 +142,8 @@ int main(int argc, char ** argv) {
         }
     }
 
+    auto * mem_tgt = llama_get_memory(ctx_tgt);
+    auto * mem_dft = llama_get_memory(ctx_dft);
 
     // Tokenize the prompt
     std::vector<llama_token> inp;
@@ -420,14 +422,14 @@ int main(int argc, char ** argv) {
             {
                 LOG_DBG("keeping sequence %d, n_past_tgt = %d, n_past_dft = %d\n", s_keep, n_past_tgt, n_past_dft);
 
-                llama_kv_self_seq_keep(ctx_dft, s_keep);
-                llama_kv_self_seq_cp  (ctx_dft, s_keep, 0, -1, -1);
-                llama_kv_self_seq_keep(ctx_dft, 0);
+                llama_memory_seq_keep(mem_dft, s_keep);
+                llama_memory_seq_cp  (mem_dft, s_keep, 0, -1, -1);
+                llama_memory_seq_keep(mem_dft, 0);
 
-                llama_kv_self_seq_rm  (ctx_tgt, s_keep, n_past_tgt, -1);
-                llama_kv_self_seq_keep(ctx_tgt, s_keep);
-                llama_kv_self_seq_cp  (ctx_tgt, s_keep, 0, -1, -1);
-                llama_kv_self_seq_keep(ctx_tgt, 0);
+                llama_memory_seq_rm  (mem_tgt, s_keep, n_past_tgt, -1);
+                llama_memory_seq_keep(mem_tgt, s_keep);
+                llama_memory_seq_cp  (mem_tgt, s_keep, 0, -1, -1);
+                llama_memory_seq_keep(mem_tgt, 0);
             }
 
             for (int s = 0; s < n_seq_dft; ++s) {
@@ -444,7 +446,7 @@ int main(int argc, char ** argv) {
             common_batch_clear(batch_dft);
             common_batch_add  (batch_dft, token_id, n_past_dft, { 0 }, true);
 
-            llama_kv_self_seq_rm(ctx_dft, 0, n_past_dft, -1);
+            llama_memory_seq_rm(mem_dft, 0, n_past_dft, -1);
             // LOG_DBG("dft batch: %s\n", LOG_BATCH_TOSTR_PRETTY(ctx_dft, batch_dft).c_str());
             llama_decode(ctx_dft, batch_dft);
 
@@ -503,8 +505,8 @@ int main(int argc, char ** argv) {
                     if (n_seq_cur < n_seq_dft && cur_p->data[f].p > p_draft_split) {
                         LOG_DBG("splitting seq %3d into %3d\n", s, n_seq_cur);
 
-                        llama_kv_self_seq_rm(ctx_dft,    n_seq_cur, -1, -1);
-                        llama_kv_self_seq_cp(ctx_dft, s, n_seq_cur, -1, -1);
+                        llama_memory_seq_rm(mem_dft,    n_seq_cur, -1, -1);
+                        llama_memory_seq_cp(mem_dft, s, n_seq_cur, -1, -1);
 
                         // all previous tokens from this branch are now also part of the new branch
                         for (int t = 0; t < batch_tgt.n_tokens; ++t) {
@@ -585,9 +587,9 @@ int main(int argc, char ** argv) {
 
         // evaluate the target model on the drafted tokens
         {
-            llama_kv_self_seq_keep(ctx_tgt, 0);
+            llama_memory_seq_keep(mem_tgt, 0);
             for (int s = 1; s < n_seq_dft; ++s) {
-                llama_kv_self_seq_cp(ctx_tgt, 0, s, -1, -1);
+                llama_memory_seq_cp(mem_tgt, 0, s, -1, -1);
             }
 
             // LOG_DBG("target batch: %s\n", LOG_BATCH_TOSTR_PRETTY(ctx_tgt, batch_tgt).c_str());

ggml/CMakeLists.txt

@@ -105,7 +105,7 @@ message(DEBUG "GGML_NATIVE_DEFAULT : ${GGML_NATIVE_DEFAULT}")
 message(DEBUG "INS_ENB             : ${INS_ENB}")
 
 option(GGML_CPU_HBM          "ggml: use memkind for CPU HBM" OFF)
-option(GGML_CPU_AARCH64      "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON)
+option(GGML_CPU_REPACK       "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON)
 option(GGML_CPU_KLEIDIAI     "ggml: use KleidiAI optimized kernels if applicable" OFF)
 option(GGML_SSE42            "ggml: enable SSE 4.2"          ${INS_ENB})
 option(GGML_AVX              "ggml: enable AVX"              ${INS_ENB})
@@ -137,7 +137,7 @@ set(GGML_CPU_ARM_ARCH        "" CACHE STRING "ggml: CPU architecture for ARM")
 set(GGML_CPU_POWERPC_CPUTYPE "" CACHE STRING "ggml: CPU type for PowerPC")
 
 
-if (WIN32)
+if (MINGW)
     set(GGML_WIN_VER "0x602" CACHE STRING   "ggml: Windows version")
 endif()
 

ggml/src/CMakeLists.txt

@@ -125,7 +125,6 @@ if (NOT MSVC)
 endif()
 
 if (MINGW)
-    # Target Windows 8 for PrefetchVirtualMemory
     add_compile_definitions(_WIN32_WINNT=${GGML_WIN_VER})
 endif()
 

ggml/src/ggml-cann/common.h

@@ -37,6 +37,7 @@
 #include <thread>
 #include <unistd.h>
 #include <functional>
+#include <optional>
 
 #include "../include/ggml-cann.h"
 #include "../include/ggml.h"
@@ -103,6 +104,9 @@ const ggml_cann_device_info& ggml_cann_info();
 void ggml_cann_set_device(int32_t device);
 int32_t ggml_cann_get_device();
 
+std::optional<std::string> get_env(const std::string& name);
+bool parse_bool(const std::string& value);
+
 /**
  * @brief Abstract base class for memory pools used by CANN.
  */
@@ -354,7 +358,8 @@ struct ggml_backend_cann_context {
         : device(device), name("CANN" + std::to_string(device)), task_queue(1024, device) {
         ggml_cann_set_device(device);
         description = aclrtGetSocName();
-        async_mode = (getenv("GGML_CANN_ASYNC_MODE") != nullptr);
+
+        bool async_mode = parse_bool(get_env("GGML_CANN_ASYNC_MODE").value_or(""));
         GGML_LOG_INFO("%s: device %d async operator submission is %s\n", __func__,
             device, async_mode ? "ON" : "OFF");
     }

ggml/src/ggml-cann/ggml-cann.cpp

@@ -31,6 +31,8 @@
 #include <mutex>
 #include <queue>
 #include <chrono>
+#include <unordered_set>
+#include <optional>
 
 #include "ggml-impl.h"
 #include "ggml-backend-impl.h"
@@ -93,6 +95,26 @@ int32_t ggml_cann_get_device() {
     return id;
 }
 
+/**
+ * @brief Get the value of the specified environment variable (name).
+ *        if not empty, return a std::string object
+ */
+std::optional<std::string> get_env(const std::string& name) {
+    const char* val = std::getenv(name.c_str());
+    if (!val) return std::nullopt;
+    std::string res = std::string(val);
+    std::transform(res.begin(), res.end(), res.begin(), ::tolower);
+    return res;
+}
+
+/**
+ * @brief Verify whether the environment variable is a valid value.
+ */
+bool parse_bool(const std::string& value) {
+    std::unordered_set<std::string> valid_values = {"on", "1", "yes", "y", "enable", "true"};
+    return valid_values.find(value) != valid_values.end();
+}
+
 /**
  * @brief Initialize the CANN device information.
  *
@@ -214,7 +236,7 @@ struct ggml_cann_pool_buf_prio : public ggml_cann_pool {
      * @param device The device ID to associate with this buffer pool.
      */
     explicit ggml_cann_pool_buf_prio(int device) : device(device) {
-        disable_clean = getenv("GGML_CANN_DISABLE_BUF_POOL_CLEAN") != nullptr;
+        disable_clean = parse_bool(get_env("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or(""));
     }
 
     /**
@@ -410,7 +432,7 @@ struct ggml_cann_pool_buf : public ggml_cann_pool {
      * @param device The device ID to associate with this buffer pool.
      */
     explicit ggml_cann_pool_buf(int device) : device(device) {
-        disable_clean = getenv("GGML_CANN_DISABLE_BUF_POOL_CLEAN") != nullptr;
+        disable_clean = parse_bool(get_env("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or(""));
     }
 
     /**
@@ -731,16 +753,18 @@ struct ggml_cann_pool_vmm : public ggml_cann_pool {
  */
 std::unique_ptr<ggml_cann_pool> ggml_backend_cann_context::new_pool_for_device(
     int device) {
-    bool disable_vmm = (getenv("GGML_CANN_DISABLE_VMM_POOL") != nullptr);
-    if (!disable_vmm && ggml_cann_info().devices[device].vmm) {
-        GGML_LOG_INFO("%s: device %d use vmm pool\n", __func__, device);
-        return std::unique_ptr<ggml_cann_pool>(new ggml_cann_pool_vmm(device));
-    }
-    bool enable_buf_prio = (getenv("GGML_CANN_ENABLE_BUF_PRIO_POOL") != nullptr);
-    if (enable_buf_prio) {
+    std::string mem_pool_type = get_env("GGML_CANN_MEM_POOL").value_or("");
+
+    if (mem_pool_type == "prio") {
         GGML_LOG_INFO("%s: device %d use buffer pool with priority queue\n", __func__, device);
         return std::unique_ptr<ggml_cann_pool>(new ggml_cann_pool_buf_prio(device));
     }
+
+    if (ggml_cann_info().devices[device].vmm && mem_pool_type != "leg") {
+        GGML_LOG_INFO("%s: device %d use vmm pool\n", __func__, device);
+        return std::unique_ptr<ggml_cann_pool>(new ggml_cann_pool_vmm(device));
+    }
+
     GGML_LOG_INFO("%s: device %d use buffer pool\n", __func__, device);
     return std::unique_ptr<ggml_cann_pool>(new ggml_cann_pool_buf(device));
 }

ggml/src/ggml-common.h

@@ -1074,6 +1074,10 @@ GGML_TABLE_BEGIN(uint32_t, iq3s_grid, 512)
     0x0f090307, 0x0f090501, 0x0f090b01, 0x0f0b0505, 0x0f0b0905, 0x0f0d0105, 0x0f0d0703, 0x0f0f0101,
 GGML_TABLE_END()
 
+GGML_TABLE_BEGIN(int8_t, kvalues_iq4nl, 16)
+    -127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113,
+GGML_TABLE_END()
+
 #define NGRID_IQ1S 2048
 #define IQ1S_DELTA 0.125f
 #define IQ1M_DELTA 0.125f

ggml/src/ggml-cpu/CMakeLists.txt

@@ -10,14 +10,14 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
     list (APPEND GGML_CPU_SOURCES
         ggml-cpu/ggml-cpu.c
         ggml-cpu/ggml-cpu.cpp
-        ggml-cpu/ggml-cpu-aarch64.cpp
-        ggml-cpu/ggml-cpu-aarch64.h
-        ggml-cpu/ggml-cpu-hbm.cpp
-        ggml-cpu/ggml-cpu-hbm.h
-        ggml-cpu/ggml-cpu-quants.c
-        ggml-cpu/ggml-cpu-quants.h
-        ggml-cpu/ggml-cpu-traits.cpp
-        ggml-cpu/ggml-cpu-traits.h
+        ggml-cpu/repack.cpp
+        ggml-cpu/repack.h
+        ggml-cpu/hbm.cpp
+        ggml-cpu/hbm.h
+        ggml-cpu/quants.c
+        ggml-cpu/quants.h
+        ggml-cpu/traits.cpp
+        ggml-cpu/traits.h
         ggml-cpu/amx/amx.cpp
         ggml-cpu/amx/amx.h
         ggml-cpu/amx/mmq.cpp
@@ -84,6 +84,11 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
 
     if (GGML_SYSTEM_ARCH STREQUAL "ARM")
         message(STATUS "ARM detected")
+        list(APPEND GGML_CPU_SOURCES
+            ggml-cpu/arch/arm/quants.c
+            ggml-cpu/arch/arm/repack.cpp
+            )
+
         if (MSVC AND NOT CMAKE_C_COMPILER_ID STREQUAL "Clang")
             message(FATAL_ERROR "MSVC is not supported for ARM, use clang")
         else()
@@ -167,6 +172,11 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "x86")
         message(STATUS "x86 detected")
+        list(APPEND GGML_CPU_SOURCES
+            ggml-cpu/arch/x86/quants.c
+            ggml-cpu/arch/x86/repack.cpp
+            )
+
         if (MSVC)
             # instruction set detection for MSVC only
             if (GGML_NATIVE)
@@ -302,7 +312,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
             # Since multiple variants of the CPU backend may be included in the same
             # build, using set_source_files_properties() to set the arch flags is not possible
             set(GGML_CPU_FEATS_NAME ${GGML_CPU_NAME}-feats)
-            add_library(${GGML_CPU_FEATS_NAME} OBJECT ggml-cpu/cpu-feats-x86.cpp)
+            add_library(${GGML_CPU_FEATS_NAME} OBJECT ggml-cpu/arch/x86/cpu-feats.cpp)
             target_include_directories(${GGML_CPU_FEATS_NAME} PRIVATE . .. ../include)
             target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE ${ARCH_DEFINITIONS})
             target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE GGML_BACKEND_DL GGML_BACKEND_BUILD GGML_BACKEND_SHARED)
@@ -311,6 +321,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "PowerPC")
         message(STATUS "PowerPC detected")
+        list(APPEND GGML_CPU_SOURCES ggml-cpu/arch/powerpc/quants.c)
         if (GGML_NATIVE)
             if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64")
                 file(READ "/proc/cpuinfo" POWER10_M)
@@ -338,6 +349,8 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "loongarch64")
         message(STATUS "loongarch64 detected")
+        list(APPEND GGML_CPU_SOURCES ggml-cpu/arch/loongarch/quants.c)
+
         list(APPEND ARCH_FLAGS -march=loongarch64)
         if (GGML_LASX)
             list(APPEND ARCH_FLAGS -mlasx)
@@ -347,6 +360,10 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "riscv64")
         message(STATUS "riscv64 detected")
+        list(APPEND GGML_CPU_SOURCES
+            ggml-cpu/arch/riscv/quants.c
+            ggml-cpu/arch/riscv/repack.cpp
+            )
         if (GGML_RVV)
             if (GGML_XTHEADVECTOR)
                 list(APPEND ARCH_FLAGS -march=rv64gc_xtheadvector -mabi=lp64d)
@@ -358,6 +375,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "s390x")
         message(STATUS "s390x detected")
+        list(APPEND GGML_CPU_SOURCES ggml-cpu/arch/s390/quants.c)
         file(READ "/proc/cpuinfo" CPUINFO_CONTENTS)
         string(REGEX REPLACE "machine[ \t\r\n]*=[ \t\r\n]*([0-9]+)" "\\1" S390X_M ${CPUINFO_CONTENTS})
 
@@ -381,12 +399,16 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
         if (GGML_VXE)
             list(APPEND ARCH_FLAGS -mvx -mzvector)
         endif()
+    elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "wasm")
+        message(STATUS "Wasm detected")
+        list (APPEND GGML_CPU_SOURCES ggml-cpu/arch/wasm/quants.c)
     else()
-        message(STATUS "Unknown architecture")
+        message(WARNING "Unknown CPU architecture. Falling back to generic implementations.")
+        list(APPEND ARCH_FLAGS -DGGML_CPU_GENERIC)
     endif()
 
-    if (GGML_CPU_AARCH64)
-        target_compile_definitions(${GGML_CPU_NAME} PRIVATE GGML_USE_CPU_AARCH64)
+    if (GGML_CPU_REPACK)
+        target_compile_definitions(${GGML_CPU_NAME} PRIVATE GGML_USE_CPU_REPACK)
     endif()
 
     if (GGML_CPU_KLEIDIAI)

ggml/src/ggml-cpu/amx/amx.cpp

@@ -5,7 +5,7 @@
 #include "ggml-backend.h"
 #include "ggml-impl.h"
 #include "ggml-cpu.h"
-#include "ggml-cpu-traits.h"
+#include "traits.h"
 
 #if defined(__gnu_linux__)
 #include <sys/syscall.h>

ggml/src/ggml-cpu/amx/mmq.cpp

@@ -8,7 +8,7 @@
 #include "mmq.h"
 #include "ggml-impl.h"
 #include "ggml-cpu-impl.h"
-#include "ggml-cpu-quants.h"
+#include "quants.h"
 #include "ggml-quants.h"
 #include <algorithm>
 #include <type_traits>

ggml/src/ggml-cpu/arch/riscv/repack.cpp

@@ -0,0 +1,396 @@
+#define GGML_COMMON_IMPL_CPP
+#define GGML_COMMON_DECL_CPP
+#include "ggml-common.h"
+#include "ggml-backend-impl.h"
+
+#include "ggml-impl.h"
+#include "ggml-cpu.h"
+#include "ggml-cpu-impl.h"
+#include "traits.h"
+
+#include <cmath>
+#include <cstring>
+#include <cassert>
+#include <cstdlib> // for qsort
+#include <cstdio>  // for GGML_ASSERT
+
+#define GGML_CPU_CLANG_WORKAROUND
+#include "../../repack.h"
+
+#if defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Woverlength-strings"
+#endif
+
+#define UNUSED GGML_UNUSED
+
+void ggml_gemv_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
+    const int qk = QK8_0;
+    const int nb = n / qk;
+    const int ncols_interleaved = 8;
+    const int blocklen = 8;
+
+    assert (n % qk == 0);
+    assert (nc % ncols_interleaved == 0);
+
+    UNUSED(s);
+    UNUSED(bs);
+    UNUSED(vx);
+    UNUSED(vy);
+    UNUSED(nr);
+    UNUSED(nc);
+    UNUSED(nb);
+    UNUSED(ncols_interleaved);
+    UNUSED(blocklen);
+
+#if defined __riscv_v
+    if (__riscv_vlenb() >= QK4_0) {
+        const size_t vl = QK4_0;
+
+        const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
+        for (int x = 0; x < nc / ncols_interleaved; x++) {
+            const block_q4_0x8 * b_ptr = (const block_q4_0x8 *) vx + (x * nb);
+
+            vfloat32m1_t sumf = __riscv_vfmv_v_f_f32m1(0.0, vl / 4);
+            for (int l = 0; l < nb; l++) {
+                const int64_t a0 = *(const int64_t *)&a_ptr[l].qs[0];
+                const int64_t a1 = *(const int64_t *)&a_ptr[l].qs[8];
+                const int64_t a2 = *(const int64_t *)&a_ptr[l].qs[16];
+                const int64_t a3 = *(const int64_t *)&a_ptr[l].qs[24];
+                __asm__ __volatile__("" ::: "memory"); // prevent gcc from emitting fused vlse64, violating alignment constraints
+                const vint8m2_t lhs_0_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(a0, vl / 4));
+                const vint8m2_t lhs_1_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(a1, vl / 4));
+                const vint8m2_t lhs_2_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(a2, vl / 4));
+                const vint8m2_t lhs_3_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(a3, vl / 4));
+
+                const vint8m4_t rhs_raw_vec = __riscv_vle8_v_i8m4((const int8_t *)b_ptr[l].qs, vl * 4);
+                const vint8m4_t rhs_vec_lo = __riscv_vsra_vx_i8m4(__riscv_vsll_vx_i8m4(rhs_raw_vec, 4, vl * 4), 4, vl * 4);
+                const vint8m4_t rhs_vec_hi = __riscv_vsra_vx_i8m4(rhs_raw_vec, 4, vl * 4);
+                const vint8m2_t rhs_vec_lo_0 = __riscv_vget_v_i8m4_i8m2(rhs_vec_lo, 0);
+                const vint8m2_t rhs_vec_lo_1 = __riscv_vget_v_i8m4_i8m2(rhs_vec_lo, 1);
+                const vint8m2_t rhs_vec_hi_0 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 0);
+                const vint8m2_t rhs_vec_hi_1 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 1);
+
+                const vint16m4_t sumi_lo_0 = __riscv_vwmul_vv_i16m4(rhs_vec_lo_0, lhs_0_8, vl * 2);
+                const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
+                const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
+                const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
+
+                const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_hi_m));
+                const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
+                const vuint16m2_t sumi_h2_1 = __riscv_vnsrl_wx_u16m2(sumi_i32, 16, vl);
+                const vuint16m2_t sumi_h2 = __riscv_vadd_vv_u16m2(sumi_h2_0, sumi_h2_1, vl);
+                const vuint32m2_t sumi_h2_i32 = __riscv_vreinterpret_v_u16m2_u32m2(sumi_h2);
+                const vuint16m1_t sumi_h4_0 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 0, vl / 2);
+                const vuint16m1_t sumi_h4_1 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 16, vl / 2);
+                const vuint16m1_t sumi_h4 = __riscv_vadd_vv_u16m1(sumi_h4_0, sumi_h4_1, vl / 2);
+                const vuint32m1_t sumi_h4_i32 = __riscv_vreinterpret_v_u16m1_u32m1(sumi_h4);
+                const vint16mf2_t sumi_h8_0 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 0, vl / 4));
+                const vint16mf2_t sumi_h8_1 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 16, vl / 4));
+                const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
+                const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);
+
+                // vector version needs Zvfhmin extension
+                const float a_scale = GGML_FP16_TO_FP32(a_ptr[l].d);
+                const float b_scales[8] = {
+                    GGML_FP16_TO_FP32(b_ptr[l].d[0]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[1]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[2]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[3]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[4]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[5]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[6]),
+                    GGML_FP16_TO_FP32(b_ptr[l].d[7])
+                };
+                const vfloat32m1_t b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
+                const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scale, vl / 4);
+                sumf = __riscv_vfmacc_vv_f32m1(sumf, tmp1, b_scales_vec, vl / 4);
+            }
+            __riscv_vse32_v_f32m1(s + x * ncols_interleaved, sumf, vl / 4);
+        }
+        return;
+    }
+
+#endif
+    {
+        float sumf[8];
+        int sumi;
+
+        const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
+        for (int x = 0; x < nc / ncols_interleaved; x++) {
+            const block_q4_0x8 * b_ptr = (const block_q4_0x8 *) vx + (x * nb);
+
+            for (int j = 0; j < ncols_interleaved; j++) sumf[j] = 0.0;
+            for (int l = 0; l < nb; l++) {
+                for (int k = 0; k < (qk / (2 * blocklen)); k++) {
+                    for (int j = 0; j < ncols_interleaved; j++) {
+                        sumi = 0;
+                        for (int i = 0; i < blocklen; ++i) {
+                            const int v0 = (int8_t) (b_ptr[l].qs[k * ncols_interleaved * blocklen + j * blocklen + i] << 4);
+                            const int v1 = (int8_t) (b_ptr[l].qs[k * ncols_interleaved * blocklen + j * blocklen + i] & 0xF0);
+                            sumi += ((v0 * a_ptr[l].qs[k * blocklen + i]) + (v1 * a_ptr[l].qs[k * blocklen + i + qk / 2])) >> 4;
+                        }
+                        sumf[j] += sumi * GGML_FP16_TO_FP32(b_ptr[l].d[j]) * GGML_FP16_TO_FP32(a_ptr[l].d);
+                    }
+                }
+            }
+            for (int j = 0; j < ncols_interleaved; j++) s[x * ncols_interleaved + j] = sumf[j];
+        }
+    }
+}
+
+void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
+    const int qk = QK8_0;
+    const int nb = n / qk;
+    const int ncols_interleaved = 8;
+    const int blocklen = 8;
+
+    assert (n % qk == 0);
+    assert (nr % 4 == 0);
+    assert (nc % ncols_interleaved == 0);
+
+    UNUSED(s);
+    UNUSED(bs);
+    UNUSED(vx);
+    UNUSED(vy);
+    UNUSED(nr);
+    UNUSED(nc);
+    UNUSED(nb);
+    UNUSED(ncols_interleaved);
+    UNUSED(blocklen);
+
+#if defined __riscv_v
+    if (__riscv_vlenb() >= QK4_0) {
+        const size_t vl = QK4_0;
+
+        for (int y = 0; y < nr / 4; y++) {
+            const block_q8_0x4 * a_ptr = (const block_q8_0x4 *) vy + (y * nb);
+            for (int x = 0; x < nc / ncols_interleaved; x++) {
+                const block_q4_0x8 * b_ptr = (const block_q4_0x8 *) vx + (x * nb);
+                vfloat32m1_t sumf0 = __riscv_vfmv_v_f_f32m1(0.0, vl / 4);
+                vfloat32m1_t sumf1 = __riscv_vfmv_v_f_f32m1(0.0, vl / 4);
+                vfloat32m1_t sumf2 = __riscv_vfmv_v_f_f32m1(0.0, vl / 4);
+                vfloat32m1_t sumf3 = __riscv_vfmv_v_f_f32m1(0.0, vl / 4);
+                for (int l = 0; l < nb; l++) {
+                    const vint8m4_t rhs_raw_vec = __riscv_vle8_v_i8m4((const int8_t *)b_ptr[l].qs, vl * 4);
+                    const vint8m4_t rhs_vec_lo = __riscv_vsra_vx_i8m4(__riscv_vsll_vx_i8m4(rhs_raw_vec, 4, vl * 4), 4, vl * 4);
+                    const vint8m4_t rhs_vec_hi = __riscv_vsra_vx_i8m4(rhs_raw_vec, 4, vl * 4);
+                    const vint8m2_t rhs_vec_lo_0 = __riscv_vget_v_i8m4_i8m2(rhs_vec_lo, 0);
+                    const vint8m2_t rhs_vec_lo_1 = __riscv_vget_v_i8m4_i8m2(rhs_vec_lo, 1);
+                    const vint8m2_t rhs_vec_hi_0 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 0);
+                    const vint8m2_t rhs_vec_hi_1 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 1);
+
+                    // vector version needs Zvfhmin extension
+                    const float a_scales[4] = {
+                        GGML_FP16_TO_FP32(a_ptr[l].d[0]),
+                        GGML_FP16_TO_FP32(a_ptr[l].d[1]),
+                        GGML_FP16_TO_FP32(a_ptr[l].d[2]),
+                        GGML_FP16_TO_FP32(a_ptr[l].d[3])
+                    };
+                    const float b_scales[8] = {
+                        GGML_FP16_TO_FP32(b_ptr[l].d[0]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[1]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[2]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[3]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[4]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[5]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[6]),
+                        GGML_FP16_TO_FP32(b_ptr[l].d[7])
+                    };
+                    const vfloat32m1_t b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
+
+                    const int64_t A0 = *(const int64_t *)&a_ptr[l].qs[0];
+                    const int64_t A4 = *(const int64_t *)&a_ptr[l].qs[32];
+                    const int64_t A8 = *(const int64_t *)&a_ptr[l].qs[64];
+                    const int64_t Ac = *(const int64_t *)&a_ptr[l].qs[96];
+                    __asm__ __volatile__("" ::: "memory"); // prevent gcc from emitting fused vlse64, violating alignment
+                    vint16m4_t sumi_l0;
+                    {
+                        const vint8m2_t lhs_0_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A0, vl / 4));
+                        const vint8m2_t lhs_1_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A4, vl / 4));
+                        const vint8m2_t lhs_2_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A8, vl / 4));
+                        const vint8m2_t lhs_3_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Ac, vl / 4));
+                        const vint16m4_t sumi_lo_0 = __riscv_vwmul_vv_i16m4(rhs_vec_lo_0, lhs_0_8, vl * 2);
+                        const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
+                        const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
+                        const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
+
+                        sumi_l0 = sumi_hi_m;
+                    }
+
+                    {
+                        const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_l0));
+                        const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
+                        const vuint16m2_t sumi_h2_1 = __riscv_vnsrl_wx_u16m2(sumi_i32, 16, vl);
+                        const vuint16m2_t sumi_h2 = __riscv_vadd_vv_u16m2(sumi_h2_0, sumi_h2_1, vl);
+                        const vuint32m2_t sumi_h2_i32 = __riscv_vreinterpret_v_u16m2_u32m2(sumi_h2);
+                        const vuint16m1_t sumi_h4_0 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 0, vl / 2);
+                        const vuint16m1_t sumi_h4_1 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 16, vl / 2);
+                        const vuint16m1_t sumi_h4 = __riscv_vadd_vv_u16m1(sumi_h4_0, sumi_h4_1, vl / 2);
+                        const vuint32m1_t sumi_h4_i32 = __riscv_vreinterpret_v_u16m1_u32m1(sumi_h4);
+                        const vint16mf2_t sumi_h8_0 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 0, vl / 4));
+                        const vint16mf2_t sumi_h8_1 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 16, vl / 4));
+                        const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
+                        const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);
+
+                        const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scales[0], vl / 4);
+                        sumf0 = __riscv_vfmacc_vv_f32m1(sumf0, tmp1, b_scales_vec, vl / 4);
+                    }
+
+                    const int64_t A1 = *(const int64_t *)&a_ptr[l].qs[8];
+                    const int64_t A5 = *(const int64_t *)&a_ptr[l].qs[40];
+                    const int64_t A9 = *(const int64_t *)&a_ptr[l].qs[72];
+                    const int64_t Ad = *(const int64_t *)&a_ptr[l].qs[104];
+                    __asm__ __volatile__("" ::: "memory"); // prevent gcc from emitting fused vlse64, violating alignment
+                    vint16m4_t sumi_l1;
+                    {
+                        const vint8m2_t lhs_0_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A1, vl / 4));
+                        const vint8m2_t lhs_1_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A5, vl / 4));
+                        const vint8m2_t lhs_2_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A9, vl / 4));
+                        const vint8m2_t lhs_3_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Ad, vl / 4));
+                        const vint16m4_t sumi_lo_0 = __riscv_vwmul_vv_i16m4(rhs_vec_lo_0, lhs_0_8, vl * 2);
+                        const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
+                        const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
+                        const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
+
+                        sumi_l1 = sumi_hi_m;
+                    }
+
+                    {
+                        const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_l1));
+                        const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
+                        const vuint16m2_t sumi_h2_1 = __riscv_vnsrl_wx_u16m2(sumi_i32, 16, vl);
+                        const vuint16m2_t sumi_h2 = __riscv_vadd_vv_u16m2(sumi_h2_0, sumi_h2_1, vl);
+                        const vuint32m2_t sumi_h2_i32 = __riscv_vreinterpret_v_u16m2_u32m2(sumi_h2);
+                        const vuint16m1_t sumi_h4_0 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 0, vl / 2);
+                        const vuint16m1_t sumi_h4_1 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 16, vl / 2);
+                        const vuint16m1_t sumi_h4 = __riscv_vadd_vv_u16m1(sumi_h4_0, sumi_h4_1, vl / 2);
+                        const vuint32m1_t sumi_h4_i32 = __riscv_vreinterpret_v_u16m1_u32m1(sumi_h4);
+                        const vint16mf2_t sumi_h8_0 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 0, vl / 4));
+                        const vint16mf2_t sumi_h8_1 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 16, vl / 4));
+                        const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
+                        const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);
+
+                        const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scales[1], vl / 4);
+                        sumf1 = __riscv_vfmacc_vv_f32m1(sumf1, tmp1, b_scales_vec, vl / 4);
+                    }
+
+                    const int64_t A2 = *(const int64_t *)&a_ptr[l].qs[16];
+                    const int64_t A6 = *(const int64_t *)&a_ptr[l].qs[48];
+                    const int64_t Aa = *(const int64_t *)&a_ptr[l].qs[80];
+                    const int64_t Ae = *(const int64_t *)&a_ptr[l].qs[112];
+                    __asm__ __volatile__("" ::: "memory"); // prevent gcc from emitting fused vlse64, violating alignment
+                    vint16m4_t sumi_l2;
+                    {
+                        const vint8m2_t lhs_0_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A2, vl / 4));
+                        const vint8m2_t lhs_1_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A6, vl / 4));
+                        const vint8m2_t lhs_2_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Aa, vl / 4));
+                        const vint8m2_t lhs_3_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Ae, vl / 4));
+                        const vint16m4_t sumi_lo_0 = __riscv_vwmul_vv_i16m4(rhs_vec_lo_0, lhs_0_8, vl * 2);
+                        const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
+                        const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
+                        const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
+
+                        sumi_l2 = sumi_hi_m;
+                    }
+
+                    {
+                        const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_l2));
+                        const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
+                        const vuint16m2_t sumi_h2_1 = __riscv_vnsrl_wx_u16m2(sumi_i32, 16, vl);
+                        const vuint16m2_t sumi_h2 = __riscv_vadd_vv_u16m2(sumi_h2_0, sumi_h2_1, vl);
+                        const vuint32m2_t sumi_h2_i32 = __riscv_vreinterpret_v_u16m2_u32m2(sumi_h2);
+                        const vuint16m1_t sumi_h4_0 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 0, vl / 2);
+                        const vuint16m1_t sumi_h4_1 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 16, vl / 2);
+                        const vuint16m1_t sumi_h4 = __riscv_vadd_vv_u16m1(sumi_h4_0, sumi_h4_1, vl / 2);
+                        const vuint32m1_t sumi_h4_i32 = __riscv_vreinterpret_v_u16m1_u32m1(sumi_h4);
+                        const vint16mf2_t sumi_h8_0 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 0, vl / 4));
+                        const vint16mf2_t sumi_h8_1 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 16, vl / 4));
+                        const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
+                        const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);
+
+                        const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scales[2], vl / 4);
+                        sumf2 = __riscv_vfmacc_vv_f32m1(sumf2, tmp1, b_scales_vec, vl / 4);
+                    }
+
+                    const int64_t A3 = *(const int64_t *)&a_ptr[l].qs[24];
+                    const int64_t A7 = *(const int64_t *)&a_ptr[l].qs[56];
+                    const int64_t Ab = *(const int64_t *)&a_ptr[l].qs[88];
+                    const int64_t Af = *(const int64_t *)&a_ptr[l].qs[120];
+                    __asm__ __volatile__("" ::: "memory"); // prevent gcc from emitting fused vlse64, violating alignment
+                    vint16m4_t sumi_l3;
+                    {
+                        const vint8m2_t lhs_0_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A3, vl / 4));
+                        const vint8m2_t lhs_1_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(A7, vl / 4));
+                        const vint8m2_t lhs_2_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Ab, vl / 4));
+                        const vint8m2_t lhs_3_8 =__riscv_vreinterpret_v_i64m2_i8m2(__riscv_vmv_v_x_i64m2(Af, vl / 4));
+                        const vint16m4_t sumi_lo_0 = __riscv_vwmul_vv_i16m4(rhs_vec_lo_0, lhs_0_8, vl * 2);
+                        const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
+                        const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
+                        const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
+
+                        sumi_l3 = sumi_hi_m;
+                    }
+
+                    {
+                        const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_l3));
+                        const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
+                        const vuint16m2_t sumi_h2_1 = __riscv_vnsrl_wx_u16m2(sumi_i32, 16, vl);
+                        const vuint16m2_t sumi_h2 = __riscv_vadd_vv_u16m2(sumi_h2_0, sumi_h2_1, vl);
+                        const vuint32m2_t sumi_h2_i32 = __riscv_vreinterpret_v_u16m2_u32m2(sumi_h2);
+                        const vuint16m1_t sumi_h4_0 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 0, vl / 2);
+                        const vuint16m1_t sumi_h4_1 = __riscv_vnsrl_wx_u16m1(sumi_h2_i32, 16, vl / 2);
+                        const vuint16m1_t sumi_h4 = __riscv_vadd_vv_u16m1(sumi_h4_0, sumi_h4_1, vl / 2);
+                        const vuint32m1_t sumi_h4_i32 = __riscv_vreinterpret_v_u16m1_u32m1(sumi_h4);
+                        const vint16mf2_t sumi_h8_0 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 0, vl / 4));
+                        const vint16mf2_t sumi_h8_1 = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vnsrl_wx_u16mf2(sumi_h4_i32, 16, vl / 4));
+                        const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
+                        const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);
+
+                        const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scales[3], vl / 4);
+                        sumf3 = __riscv_vfmacc_vv_f32m1(sumf3, tmp1, b_scales_vec, vl / 4);
+                    }
+                }
+                __riscv_vse32_v_f32m1(&s[(y * 4 + 0) * bs + x * ncols_interleaved], sumf0, vl / 4);
+                __riscv_vse32_v_f32m1(&s[(y * 4 + 1) * bs + x * ncols_interleaved], sumf1, vl / 4);
+                __riscv_vse32_v_f32m1(&s[(y * 4 + 2) * bs + x * ncols_interleaved], sumf2, vl / 4);
+                __riscv_vse32_v_f32m1(&s[(y * 4 + 3) * bs + x * ncols_interleaved], sumf3, vl / 4);
+            }
+        }
+
+        return;
+    }
+
+#endif // #if ! ((defined(_MSC_VER)) && ! defined(__clang__)) && defined(__aarch64__)
+    float sumf[4][8];
+    int sumi;
+
+    for (int y = 0; y < nr / 4; y++) {
+        const block_q8_0x4 * a_ptr = (const block_q8_0x4 *) vy + (y * nb);
+        for (int x = 0; x < nc / ncols_interleaved; x++) {
+            const block_q4_0x8 * b_ptr = (const block_q4_0x8 *) vx + (x * nb);
+            for (int m = 0; m < 4; m++) {
+                for (int j = 0; j < ncols_interleaved; j++) sumf[m][j] = 0.0;
+            }
+            for (int l = 0; l < nb; l++) {
+                for (int k = 0; k < (qk / (2 * blocklen)); k++) {
+                    for (int m = 0; m < 4; m++) {
+                        for (int j = 0; j < ncols_interleaved; j++) {
+                            sumi = 0;
+                            for (int i = 0; i < blocklen; ++i) {
+                                const int v0 = (int8_t) (b_ptr[l].qs[k * ncols_interleaved * blocklen + j * blocklen + i] << 4);
+                                const int v1 = (int8_t) (b_ptr[l].qs[k * ncols_interleaved * blocklen + j * blocklen + i] & 0xF0);
+                                sumi += ((v0 * a_ptr[l].qs[k * 4 * blocklen + m * blocklen + i]) +
+                                         (v1 * a_ptr[l].qs[k * 4 * blocklen + m * blocklen + i + qk / 2 * 4])) >> 4;
+                            }
+                            sumf[m][j] += sumi * GGML_FP16_TO_FP32(b_ptr[l].d[j]) * GGML_FP16_TO_FP32(a_ptr[l].d[m]);
+                        }
+                    }
+                }
+            }
+            for (int m = 0; m < 4; m++) {
+                for (int j = 0; j < ncols_interleaved; j++)
+                    s[(y * 4 + m) * bs + x * ncols_interleaved + j] = sumf[m][j];
+            }
+        }
+    }
+}

ggml/src/ggml-cpu/common.h

@@ -1,7 +1,7 @@
 #pragma once
 
 #include "ggml.h"
-#include "ggml-cpu-traits.h"
+#include "traits.h"
 #include "ggml-cpu-impl.h"
 #include "ggml-impl.h"
 

ggml/src/ggml-cpu/ggml-cpu-aarch64.h

@@ -1,8 +0,0 @@
-#pragma once
-
-#include "ggml-cpu-traits.h"
-#include "ggml.h"
-
-// GGML internal header
-
-ggml_backend_buffer_type_t ggml_backend_cpu_aarch64_buffer_type(void);

ggml/src/ggml-cpu/ggml-cpu-impl.h

@@ -506,3 +506,25 @@ void ggml_barrier(struct ggml_threadpool * tp);
 #ifdef __cplusplus
 }
 #endif
+
+#define GGML_DO_PRAGMA_(x) _Pragma (#x)
+#define GGML_DO_PRAGMA(x) GGML_DO_PRAGMA_(x)
+#if defined(GGML_CPU_GENERIC) || defined(__HIPCC__)
+// Note for Apple targets:
+// - clang: aliases are not supported on darwin
+// - all native kernels need to be implemented in both x86 and arm files
+// - on iOS, tvOS, and visionOS, if cmake cannot determine the target architecture, all `_generic` names are replaced by defines
+# define GGML_WEAK_ALIAS(name, alias)
+#elif defined(__GNUC__)
+// GCC/Clang on *nix
+# define GGML_WEAK_ALIAS(name, alias) GGML_DO_PRAGMA(weak name = alias) // NOLINT
+#elif defined(_MSC_VER) && defined (_WIN64)
+// MSVC
+// Note: C name mangling varies across different calling conventions
+// see https://learn.microsoft.com/en-us/cpp/build/reference/decorated-names?view=msvc-170
+# define GGML_WEAK_ALIAS(name, alias) GGML_DO_PRAGMA(comment(linker, "/alternatename:" #name "=" #alias))
+#else
+# error "Unsupported compiler for GGML_WEAK_ALIAS"
+#endif
+
+#define GGML_CPU_NATIVE_IMPL(name) GGML_WEAK_ALIAS(name, name ## _generic)

ggml/src/ggml-cpu/ggml-cpu-quants.h

@@ -1,63 +0,0 @@
-#pragma once
-
-#define GGML_COMMON_DECL_C
-#include "ggml-common.h"
-
-#include "ggml.h"
-
-// GGML CPU internal header
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-// Quantization
-void quantize_row_q4_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q4_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-
-void quantize_row_q2_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q3_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q4_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q6_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-
-void quantize_row_tq1_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_tq2_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-
-void quantize_row_iq4_nl (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-void quantize_row_iq4_xs (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
-
-// Dot product
-void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-
-void ggml_vec_dot_q2_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q3_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q4_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q5_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-
-void ggml_vec_dot_tq1_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_tq2_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-
-void ggml_vec_dot_iq2_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq2_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq2_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq3_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq1_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq1_m_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_iq3_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-
-#ifdef __cplusplus
-}
-#endif

ggml/src/ggml-cpu/ggml-cpu.c

@@ -3,11 +3,11 @@
 
 #include "ggml-backend-impl.h"
 #include "ggml-backend.h"
-#include "ggml-cpu-traits.h"
+#include "traits.h"
 #include "ggml-cpu-impl.h"
 #include "ggml-cpu.h"
 #include "ggml-impl.h"
-#include "ggml-cpu-quants.h"
+#include "quants.h"
 #include "ggml-threading.h"
 #include "unary-ops.h"
 #include "binary-ops.h"

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -1,8 +1,8 @@
 #include "ggml-backend.h"
 #include "ggml-backend-impl.h"
 #include "ggml-cpu.h"
-#include "ggml-cpu-aarch64.h"
-#include "ggml-cpu-traits.h"
+#include "repack.h"
+#include "traits.h"
 #include "ggml-impl.h"
 #include "amx/amx.h"
 
@@ -11,7 +11,7 @@
 #include <vector>
 
 #ifdef GGML_USE_CPU_HBM
-#    include "ggml-cpu-hbm.h"
+#    include "hbm.h"
 #endif
 
 #ifdef GGML_USE_CPU_KLEIDIAI
@@ -51,9 +51,9 @@ std::vector<ggml_backend_buffer_type_t>& ggml_backend_cpu_get_extra_buffers_type
         }
 #endif
 
-#ifdef GGML_USE_CPU_AARCH64
-        if (ggml_backend_cpu_aarch64_buffer_type()) {
-            bufts.push_back(ggml_backend_cpu_aarch64_buffer_type());
+#ifdef GGML_USE_CPU_REPACK
+        if (ggml_backend_cpu_repack_buffer_type()) {
+            bufts.push_back(ggml_backend_cpu_repack_buffer_type());
         }
 #endif
 
@@ -596,8 +596,8 @@ static ggml_backend_feature * ggml_backend_cpu_get_features(ggml_backend_reg_t r
     #ifdef GGML_USE_CPU_KLEIDIAI
         features.push_back({ "KLEIDIAI", "1" });
     #endif
-    #ifdef GGML_USE_CPU_AARCH64
-        features.push_back({ "AARCH64_REPACK", "1" });
+    #ifdef GGML_USE_CPU_REPACK
+        features.push_back({ "REPACK", "1" });
     #endif
 
         features.push_back({ nullptr, nullptr });

ggml/src/ggml-cpu/hbm.cpp

@@ -5,7 +5,7 @@
 #include "ggml-cpu.h"
 #include "ggml-impl.h"
 
-#include "ggml-cpu-hbm.h"
+#include "hbm.h"
 
 // buffer type HBM
 

ggml/src/ggml-cpu/kleidiai/kleidiai.cpp

@@ -26,7 +26,7 @@
 #include "ggml-impl.h"
 #include "ggml-backend-impl.h"
 #include "ggml-threading.h"
-#include "ggml-cpu-traits.h"
+#include "traits.h"
 
 #include "kernels.h"
 

ggml/src/ggml-cpu/ops.cpp

@@ -8132,8 +8132,8 @@ static void ggml_compute_forward_rwkv_wkv6_f32(
         #define WKV_VECTOR_SIZE 4
     #endif
 
-    int wkv_vector_size;
     #ifdef WKV_VECTOR_SIZE
+        int wkv_vector_size;
         #if defined(__ARM_FEATURE_SVE)
             wkv_vector_size = svcntw();
         #else
@@ -8348,8 +8348,8 @@ static void ggml_compute_forward_gla_f32(
         #define GLA_VECTOR_SIZE 4
     #endif
 
-    int gla_vector_size;
     #ifdef GLA_VECTOR_SIZE
+        int gla_vector_size;
         #if defined(__ARM_FEATURE_SVE)
             gla_vector_size = svcntw();
         #else

ggml/src/ggml-cpu/quants.h

@@ -0,0 +1,116 @@
+#pragma once
+
+#define GGML_COMMON_DECL_C
+#include "ggml-common.h"
+
+#include "ggml.h"
+
+// GGML CPU internal header
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Quantization
+void quantize_row_q4_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_q2_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q3_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q6_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_tq1_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_tq2_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_iq4_nl (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq4_xs (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+// Dot product
+void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+void ggml_vec_dot_q2_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q3_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+void ggml_vec_dot_tq1_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_tq2_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+void ggml_vec_dot_iq2_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_m_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+// Generic implementation
+void quantize_row_q8_0_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void quantize_row_q8_1_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void quantize_row_q8_K_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void ggml_vec_dot_q4_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_1_q8_1_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_1_q8_1_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q8_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_tq1_0_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_tq2_0_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q2_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q3_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy,  size_t by, int nrc);
+void ggml_vec_dot_q6_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_xxs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_xs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_xxs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_m_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_nl_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_xs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+#if defined(GGML_CPU_GENERIC)
+#define quantize_row_q8_0_generic quantize_row_q8_0
+#define quantize_row_q8_1_generic quantize_row_q8_1
+#define quantize_row_q8_K_generic quantize_row_q8_K
+#define ggml_vec_dot_q4_0_q8_0_generic ggml_vec_dot_q4_0_q8_0
+#define ggml_vec_dot_q4_1_q8_1_generic ggml_vec_dot_q4_1_q8_1
+#define ggml_vec_dot_q5_0_q8_0_generic ggml_vec_dot_q5_0_q8_0
+#define ggml_vec_dot_q5_1_q8_1_generic ggml_vec_dot_q5_1_q8_1
+#define ggml_vec_dot_q8_0_q8_0_generic ggml_vec_dot_q8_0_q8_0
+#define ggml_vec_dot_tq1_0_q8_K_generic ggml_vec_dot_tq1_0_q8_K
+#define ggml_vec_dot_tq2_0_q8_K_generic ggml_vec_dot_tq2_0_q8_K
+#define ggml_vec_dot_q2_K_q8_K_generic ggml_vec_dot_q2_K_q8_K
+#define ggml_vec_dot_q3_K_q8_K_generic ggml_vec_dot_q3_K_q8_K
+#define ggml_vec_dot_q4_K_q8_K_generic ggml_vec_dot_q4_K_q8_K
+#define ggml_vec_dot_q5_K_q8_K_generic ggml_vec_dot_q5_K_q8_K
+#define ggml_vec_dot_q6_K_q8_K_generic ggml_vec_dot_q6_K_q8_K
+#define ggml_vec_dot_iq2_xxs_q8_K_generic ggml_vec_dot_iq2_xxs_q8_K
+#define ggml_vec_dot_iq2_xs_q8_K_generic ggml_vec_dot_iq2_xs_q8_K
+#define ggml_vec_dot_iq2_s_q8_K_generic ggml_vec_dot_iq2_s_q8_K
+#define ggml_vec_dot_iq3_xxs_q8_K_generic ggml_vec_dot_iq3_xxs_q8_K
+#define ggml_vec_dot_iq3_s_q8_K_generic ggml_vec_dot_iq3_s_q8_K
+#define ggml_vec_dot_iq1_s_q8_K_generic ggml_vec_dot_iq1_s_q8_K
+#define ggml_vec_dot_iq1_m_q8_K_generic ggml_vec_dot_iq1_m_q8_K
+#define ggml_vec_dot_iq4_nl_q8_0_generic ggml_vec_dot_iq4_nl_q8_0
+#define ggml_vec_dot_iq4_xs_q8_K_generic ggml_vec_dot_iq4_xs_q8_K
+#endif
+
+#ifdef __cplusplus
+}
+#endif

ggml/src/ggml-cpu/repack.h

@@ -0,0 +1,119 @@
+#pragma once
+
+#define GGML_COMMON_DECL_CPP
+#include "ggml-common.h"
+
+#include "traits.h"
+#include "ggml.h"
+
+// GGML internal header
+
+ggml_backend_buffer_type_t ggml_backend_cpu_repack_buffer_type(void);
+
+template <int K> constexpr int QK_0() {
+    if constexpr (K == 4) {
+        return QK4_0;
+    }
+    if constexpr (K == 8) {
+        return QK8_0;
+    }
+    return -1;
+}
+
+template <int K, int N> struct block {
+    ggml_half d[N];                         // deltas for N qK_0 blocks
+    int8_t    qs[(QK_0<K>() * N * K) / 8];  // quants for N qK_0 blocks
+};
+
+// control size
+static_assert(sizeof(block<4, 4>) == 4 * sizeof(ggml_half) + QK8_0 * 2, "wrong block<4,4> size/padding");
+static_assert(sizeof(block<4, 8>) == 8 * sizeof(ggml_half) + QK8_0 * 4, "wrong block<4,8> size/padding");
+static_assert(sizeof(block<8, 4>) == 4 * sizeof(ggml_half) + QK8_0 * 4, "wrong block<8,4> size/padding");
+static_assert(sizeof(block<8, 8>) == 8 * sizeof(ggml_half) + QK8_0 * 8, "wrong block<8,8> size/padding");
+
+using block_q4_0x4 = block<4, 4>;
+using block_q4_0x8 = block<4, 8>;
+using block_q8_0x4 = block<8, 4>;
+using block_q8_0x8 = block<8, 8>;
+
+struct block_q4_Kx8 {
+    ggml_half d[8];      // super-block scale for quantized scales
+    ggml_half dmin[8];   // super-block scale for quantized mins
+    uint8_t scales[96];  // scales and mins, quantized with 6 bits
+    uint8_t qs[1024];    // 4--bit quants
+};
+
+static_assert(sizeof(block_q4_Kx8) == sizeof(ggml_half) * 16 + K_SCALE_SIZE * 8 + QK_K * 4, "wrong q4_K block size/padding");
+
+struct block_q8_Kx4 {
+    float d[4];              // delta
+    int8_t qs[QK_K * 4];     // quants
+    int16_t bsums[QK_K / 4]; // sum of quants in groups of 16
+};
+
+static_assert(sizeof(block_q8_Kx4) == sizeof(float) * 4 + QK_K * 4 + (QK_K / 4) * sizeof(int16_t), "wrong q8_K block size/padding");
+
+struct block_iq4_nlx4 {
+    ggml_half d[4];            // deltas for 4 iq4_nl blocks
+    uint8_t   qs[QK4_NL * 2];  // nibbles / quants for 4 iq4_nl blocks
+};
+
+static_assert(sizeof(block_iq4_nlx4) == 4 * sizeof(ggml_half) + QK4_NL * 2, "wrong iq4_nlx4 block size/padding");
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+// Workaround for clang:
+// clang++ complains: ``error: call to 'ggml_gemm_q4_0_4x4_q8_0' is ambiguous''
+// repro: https://godbolt.org/z/oKdeWKonM (ICE), https://godbolt.org/z/1szq6P36v (ambiguous call)
+#if defined(GGML_CPU_CLANG_WORKAROUND) || !(defined(__GNUC__) && defined(__clang__)) || defined(__HIPCC__)
+void ggml_quantize_mat_q8_0_4x4(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_quantize_mat_q8_0_4x8(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_quantize_mat_q8_K_4x8(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_gemv_q4_0_4x4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_0_4x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_K_8x8_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_iq4_nl_4x4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_4x4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_4x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_K_8x8_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_iq4_nl_4x4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+#endif // !defined(__clang__)
+
+// Native implementations
+void ggml_quantize_mat_q8_0_4x4_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_quantize_mat_q8_0_4x8_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_quantize_mat_q8_K_4x8_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
+void ggml_gemv_q4_0_4x4_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_0_4x8_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_0_8x8_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_q4_K_8x8_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemv_iq4_nl_4x4_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_4x4_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_4x8_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_0_8x8_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_q4_K_8x8_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+void ggml_gemm_iq4_nl_4x4_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc);
+
+#if defined(GGML_CPU_GENERIC)
+#define ggml_quantize_mat_q8_0_4x4_generic ggml_quantize_mat_q8_0_4x4
+#define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8
+#define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8
+#define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0
+#define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0
+#define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0
+#define ggml_gemv_q4_K_8x8_q8_K_generic ggml_gemv_q4_K_8x8_q8_K
+#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
+#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
+#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
+#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
+#define ggml_gemm_q4_K_8x8_q8_K_generic ggml_gemm_q4_K_8x8_q8_K
+#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
+#endif
+
+#if defined(__cplusplus)
+} // extern "C"
+#endif

ggml/src/ggml-cpu/traits.cpp

@@ -1,4 +1,4 @@
-#include "ggml-cpu-traits.h"
+#include "traits.h"
 
 #include "ggml-backend-impl.h"
 #include "ggml-backend.h"

ggml/src/ggml-cuda/common.cuh

@@ -466,9 +466,6 @@ static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, i
 #endif // defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)
 }
 
-// TODO: move to ggml-common.h
-static constexpr __device__ int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
-
 typedef void (*dequantize_kernel_t)(const void * vx, const int64_t ib, const int iqs, dfloat2 & v);
 
 static __device__ __forceinline__ float get_alibi_slope(

ggml/src/ggml-cuda/fattn-mma-f16.cuh

@@ -652,9 +652,12 @@ static __device__ __forceinline__ void flash_attn_ext_f16_iter(
         float KQ_max_scale[cols_per_thread];
 #pragma unroll
         for (int col = 0; col < cols_per_thread; ++col) {
-            KQ_max_scale[col] = expf(KQ_max[col] - KQ_max_new[col]);
+            const float KQ_max_diff = KQ_max[col] - KQ_max_new[col];
+            KQ_max_scale[col] = expf(KQ_max_diff);
             KQ_max[col] = KQ_max_new[col];
 
+            *((uint32_t *) &KQ_max_scale[col]) *= KQ_max_diff >= SOFTMAX_FTZ_THRESHOLD;
+
             // Scale previous KQ_rowsum to account for a potential increase in KQ_max:
             KQ_rowsum[col] = KQ_max_scale[col]*KQ_rowsum[col] + KQ_rowsum_add[col];
         }

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -615,9 +615,8 @@ static void ggml_backend_cuda_buffer_clear(ggml_backend_buffer_t buffer, uint8_t
     ggml_backend_cuda_buffer_context * ctx = (ggml_backend_cuda_buffer_context *)buffer->context;
 
     ggml_cuda_set_device(ctx->device);
-    CUDA_CHECK(cudaDeviceSynchronize());
-    CUDA_CHECK(cudaMemset(ctx->dev_ptr, value, buffer->size));
-    CUDA_CHECK(cudaDeviceSynchronize());
+    CUDA_CHECK(cudaMemsetAsync(ctx->dev_ptr, value, buffer->size, cudaStreamPerThread));
+    CUDA_CHECK(cudaStreamSynchronize(cudaStreamPerThread));
 }
 
 static const ggml_backend_buffer_i ggml_backend_cuda_buffer_interface = {
@@ -1144,7 +1143,6 @@ typedef void (*ggml_cuda_op_mul_mat_t)(
 static cudaError_t ggml_cuda_cpy_tensor_2d(
     void * dst, const struct ggml_tensor * src, int64_t i3, int64_t i2, int64_t i1_low, int64_t i1_high, cudaStream_t stream) {
 
-    GGML_ASSERT(ggml_backend_buffer_is_cuda(src->buffer));
     const char * src_ptr = (const char *) src->data;
     char       * dst_ptr = (char       *) dst;
 
@@ -1427,8 +1425,6 @@ static void ggml_cuda_op_mul_mat(
     const int64_t nb2 = dst->nb[2];
     const int64_t nb3 = dst->nb[3];
 
-    GGML_ASSERT(ggml_backend_buffer_is_cuda(dst->buffer));
-    GGML_ASSERT(ggml_backend_buffer_is_cuda(src1->buffer));
     ggml_backend_cuda_buffer_context * src1_ctx = (ggml_backend_cuda_buffer_context *) src1->buffer->context;
     ggml_backend_cuda_buffer_context * dst_ctx  = (ggml_backend_cuda_buffer_context *) dst->buffer->context;
 
@@ -1750,7 +1746,7 @@ static void ggml_cuda_mul_mat_batched_cublas(ggml_backend_cuda_context & ctx, co
     GGML_ASSERT(!ggml_is_transposed(src0));
     GGML_ASSERT(!ggml_is_transposed(src1));
 
-    GGML_ASSERT(ggml_backend_buffer_is_cuda(src0->buffer));
+    GGML_ASSERT(!ggml_backend_buft_is_cuda_split(src0->buffer->buft));
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
 
     // Byte offsets and tensor dimensions are currently used in an inconsistent way for dst.

ggml/src/ggml-quants.c

@@ -2425,8 +2425,6 @@ void dequantize_row_iq1_m(const block_iq1_m * GGML_RESTRICT x, float * GGML_REST
     }
 }
 
-static const int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
-
 void dequantize_row_iq4_nl(const block_iq4_nl * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k) {
     assert(k % QK4_NL == 0);
     const int64_t nb = k / QK4_NL;

ggml/src/ggml-sycl/common.hpp

@@ -149,8 +149,6 @@ typedef sycl::float2 dfloat2;
 
 #define MMVQ_MAX_BATCH_SIZE  8
 
-static const int8_t kvalues_iq4nl[16]={-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
-
 static int g_all_sycl_device_count = -1;
 static bool g_ggml_backend_sycl_buffer_type_initialized = false;
 

ggml/src/ggml-sycl/convert.cpp

@@ -265,6 +265,17 @@ static void dequantize_row_q6_K_sycl(const void *vx, dst_t *y, const int64_t k,
 #endif
 }
 
+template <typename dst_t>
+static void dequantize_row_q6_K_sycl_reorder(const void * vx, dst_t * y, const int64_t k, dpct::queue_ptr stream) {
+    const int64_t nb = k / QK_K;
+
+    dpct::has_capability_or_fail(stream->get_device(), { sycl::aspect::fp16 });
+
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, nb) * sycl::range<3>(1, 1, 64), sycl::range<3>(1, 1, 64)),
+        [=](sycl::nd_item<3> item_ct1) { dequantize_block_q6_K_reorder(vx, y, item_ct1, nb); });
+}
+
 template <typename dst_t>
 static void dequantize_row_iq1_s_sycl(const void *vx, dst_t *y, const int64_t k,
                                         dpct::queue_ptr stream) {
@@ -530,7 +541,11 @@ to_fp16_sycl_t ggml_get_to_fp16_sycl(ggml_type type, ggml_tensor * dst) {
         case GGML_TYPE_Q5_K:
             return dequantize_row_q5_K_sycl;
         case GGML_TYPE_Q6_K:
-            return dequantize_row_q6_K_sycl;
+            if (dst->src[0]->extra && ((ggml_tensor_extra_gpu *) dst->src[0]->extra)->optimized_feature.reorder) {
+                return dequantize_row_q6_K_sycl_reorder;
+            } else {
+                return dequantize_row_q6_K_sycl;
+            }
         case GGML_TYPE_IQ1_S:
             return dequantize_row_iq1_s_sycl;
         case GGML_TYPE_IQ1_M:
@@ -587,7 +602,11 @@ to_fp32_sycl_t ggml_get_to_fp32_sycl(ggml_type type, ggml_tensor *dst) {
         case GGML_TYPE_Q5_K:
             return dequantize_row_q5_K_sycl;
         case GGML_TYPE_Q6_K:
-            return dequantize_row_q6_K_sycl;
+            if (dst->src[0]->extra && ((ggml_tensor_extra_gpu *) dst->src[0]->extra)->optimized_feature.reorder) {
+                return dequantize_row_q6_K_sycl_reorder;
+            } else {
+                return dequantize_row_q6_K_sycl;
+            }
         case GGML_TYPE_IQ1_S:
             return dequantize_row_iq1_s_sycl;
         case GGML_TYPE_IQ1_M:

ggml/src/ggml-sycl/cpy.cpp

@@ -1,8 +1,12 @@
 #include "cpy.hpp"
 
 #include <float.h>
+#include <string>
 
 #include "dequantize.hpp"
+#include "ggml-sycl/common.hpp"
+#include "ggml-sycl/presets.hpp"
+#include "ggml.h"
 
 static __dpct_inline__ int best_index_int8(int n, const int8_t * val, float x) {
     if (x <= val[0]) {
@@ -116,6 +120,15 @@ static void cpy_blck_f32_q8_0(const char * cxi, char * cdsti) {
     }
 }
 
+/* quantized type same copy */
+template<typename T>
+static void cpy_blck_q_q(const char * cxi, char * cdsti) {
+    const T * xi = (const T *) cxi;
+    T * dsti = (T *) cdsti;
+    *dsti = *xi;
+}
+
+
 static void cpy_blck_q8_0_f32(const char * cxi, char * cdsti) {
     float * cdstf = (float *) (cdsti);
 
@@ -311,6 +324,34 @@ template <dequantize_kernel_t dequant, int qk> static void cpy_blck_q_f32(const
     }
 }
 
+
+template <typename T, int qk>
+static void cpy_q_q(const char * cx, char * cdst, const int ne, const int ne00, const int ne01, const int ne02,
+                      const int nb00, const int nb01, const int nb02, const int nb03, const int ne10, const int ne11,
+                      const int ne12, const int nb10, const int nb11, const int nb12, const int nb13,
+                      const sycl::nd_item<3> & item_ct1) {
+    const int i = (item_ct1.get_local_range(2) * item_ct1.get_group(2) + item_ct1.get_local_id(2)) * qk;
+
+    if (i >= ne) {
+        return;
+    }
+
+    const int i03      = i / (ne00 * ne01 * ne02);
+    const int i02      = (i - i03 * ne00 * ne01 * ne02) / (ne00 * ne01);
+    const int i01      = (i - i03 * ne00 * ne01 * ne02 - i02 * ne01 * ne00) / ne00;
+    const int i00      = i - i03 * ne00 * ne01 * ne02 - i02 * ne01 * ne00 - i01 * ne00;
+    const int x_offset = (i00 / qk) * nb00 + i01 * nb01 + i02 * nb02 + i03 * nb03;
+
+
+    const int i13        = i / (ne10 * ne11 * ne12);
+    const int i12        = (i - i13 * ne10 * ne11 * ne12) / (ne10 * ne11);
+    const int i11        = (i - i13 * ne10 * ne11 * ne12 - i12 * ne10 * ne11) / ne10;
+    const int i10        = i - i13 * ne10 * ne11 * ne12 - i12 * ne10 * ne11 - i11 * ne10;
+    const int dst_offset = (i10 / qk) * nb10 + i11 * nb11 + i12 * nb12 + i13 * nb13;
+
+    cpy_blck_q_q<T>(cx + x_offset, cdst + dst_offset);
+}
+
 template <cpy_kernel_t cpy_blck, int qk>
 static void cpy_f32_q(const char * cx, char * cdst, const int ne, const int ne00, const int ne01, const int ne02,
                       const int nb00, const int nb01, const int nb02, const int nb03, const int ne10, const int ne11,
@@ -322,6 +363,7 @@ static void cpy_f32_q(const char * cx, char * cdst, const int ne, const int ne00
         return;
     }
 
+
     const int i03      = i / (ne00 * ne01 * ne02);
     const int i02      = (i - i03 * ne00 * ne01 * ne02) / (ne00 * ne01);
     const int i01      = (i - i03 * ne00 * ne01 * ne02 - i02 * ne01 * ne00) / ne00;
@@ -615,6 +657,70 @@ static void ggml_cpy_i32_i32_sycl(const char * cx, char * cdst, const int ne, co
     }
 }
 
+static void ggml_cpy_q8_0_q8_0(const char * cx, char * cdst, const int ne, const int ne00, const int ne01,
+                                   const int ne02, const int nb00, const int nb01, const int nb02, const int nb03,
+                                   const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                   const int nb12, const int nb13, queue_ptr stream) {
+    const int num_blocks = ceil_div(ne, SYCL_CPY_BLOCK_SIZE);
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE),
+                              sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE)), [=](sycl::nd_item<3> item_ct1) {
+            cpy_q_q<block_q8_0, QK8_0>(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, item_ct1);
+        });
+}
+
+
+static void ggml_cpy_q5_0_q5_0(const char * cx, char * cdst, const int ne, const int ne00, const int ne01,
+                                   const int ne02, const int nb00, const int nb01, const int nb02, const int nb03,
+                                   const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                   const int nb12, const int nb13, queue_ptr stream) {
+    const int num_blocks = ceil_div(ne, SYCL_CPY_BLOCK_SIZE);
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE),
+                              sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE)), [=](sycl::nd_item<3> item_ct1) {
+            cpy_q_q<block_q5_0, QK5_0>(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, item_ct1);
+        });
+}
+
+
+static void ggml_cpy_q5_1_q5_1(const char * cx, char * cdst, const int ne, const int ne00, const int ne01,
+                                   const int ne02, const int nb00, const int nb01, const int nb02, const int nb03,
+                                   const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                   const int nb12, const int nb13, queue_ptr stream) {
+    const int num_blocks = ceil_div(ne, SYCL_CPY_BLOCK_SIZE);
+
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE),
+                              sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE)), [=](sycl::nd_item<3> item_ct1) {
+            cpy_q_q<block_q5_1, QK5_1>(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, item_ct1);
+        });
+}
+
+
+static void ggml_cpy_q4_0_q4_0(const char * cx, char * cdst, const int ne, const int ne00, const int ne01,
+                                   const int ne02, const int nb00, const int nb01, const int nb02, const int nb03,
+                                   const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                   const int nb12, const int nb13, queue_ptr stream) {
+    const int num_blocks = ceil_div(ne, SYCL_CPY_BLOCK_SIZE);
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE), sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE)), [=](sycl::nd_item<3> item_ct1) {
+            cpy_q_q<block_q4_0, QK4_0>(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, item_ct1);
+        });
+}
+
+
+static void ggml_cpy_q4_1_q4_1(const char * cx, char * cdst, const int ne, const int ne00, const int ne01,
+                                   const int ne02, const int nb00, const int nb01, const int nb02, const int nb03,
+                                   const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                   const int nb12, const int nb13, queue_ptr stream) {
+
+   const int num_blocks = ceil_div(ne, SYCL_CPY_BLOCK_SIZE);
+   stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE), sycl::range<3>(1, 1, SYCL_CPY_BLOCK_SIZE)), [=](sycl::nd_item<3> item_ct1) {
+            cpy_q_q<block_q4_1, QK4_1>(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, item_ct1);
+        });
+}
+
 void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1) try {
     // Unlike other operators ggml_sycl_cpy takes 2 distinct tensors instead of a dst ggml_tensor and rely on its src field
     scope_op_debug_print scope_dbg_print(__func__, src1, /*num_src=*/0,
@@ -632,8 +738,10 @@ void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, co
 
     char * src0_ddc = (char *) src0->data;
     char * src1_ddc = (char *) src1->data;
-
-    if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32) {
+    if ((src0->type == src1->type) && (ggml_is_contiguous(src0) && ggml_is_contiguous(src1))) {
+        GGML_SYCL_DEBUG("%s: memcpy path\n", __func__);
+        main_stream->memcpy(src1_ddc, src0_ddc, ggml_nbytes(src0));
+    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32) {
         ggml_cpy_f32_f32_sycl(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10,
                               nb11, nb12, nb13, main_stream);
     } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F16) {
@@ -684,6 +792,16 @@ void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, co
     } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_IQ4_NL) {
         ggml_cpy_f32_iq4_nl_sycl(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12,
                                  nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q8_0 && src1->type == GGML_TYPE_Q8_0) {
+        ggml_cpy_q8_0_q8_0(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q5_0 && src1->type == GGML_TYPE_Q5_0) {
+        ggml_cpy_q5_0_q5_0(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q5_1 && src1->type == GGML_TYPE_Q5_1) {
+        ggml_cpy_q5_1_q5_1(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q4_0 && src1->type == GGML_TYPE_Q4_0) {
+        ggml_cpy_q4_0_q4_0(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q4_1 && src1->type == GGML_TYPE_Q4_1) {
+        ggml_cpy_q4_1_q4_1(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
     } else {
         GGML_LOG_ERROR("%s: unsupported type combination (%s to %s)\n", __func__, ggml_type_name(src0->type),
                        ggml_type_name(src1->type));

ggml/src/ggml-sycl/dequantize.hpp

@@ -538,6 +538,38 @@ static void dequantize_block_q6_K(const void * __restrict__ vx, dst_t * __restri
 #endif
 }
 
+template <typename dst_t>
+static void dequantize_block_q6_K_reorder(const void * __restrict__ vx, dst_t * __restrict__ yy,
+                                          const sycl::nd_item<3> & item_ct1, int64_t n_blocks) {
+    const int64_t ib = item_ct1.get_group(2);
+
+    const int64_t tid = item_ct1.get_local_id(2);
+    const int64_t ip  = tid / 32;       // ip is 0 or 1
+    const int64_t il  = tid - 32 * ip;  // 0...32
+    const int64_t is  = 8 * ip + il / 16;
+
+    const uint8_t *   base_ptr           = static_cast<const uint8_t *>(vx);
+    const auto        ql_offset          = ib * (QK_K / 2);
+    const auto        qh_offset          = (QK_K / 2) * n_blocks + (QK_K / 4) * ib;
+    const auto        base_scales_offset = (QK_K / 2) * n_blocks + (QK_K / 4) * n_blocks + (QK_K / 16) * ib;
+    const auto        base_d_offset      = ((QK_K / 2) + (QK_K / 4) + (QK_K / 16)) * n_blocks;
+    const uint8_t *   ql_ptr             = base_ptr + ql_offset;
+    const uint8_t *   qh_ptr             = base_ptr + qh_offset;
+    const uint8_t *   scales_ptr         = base_ptr + base_scales_offset;
+    const ggml_half * d                  = (const ggml_half *) (base_ptr + base_d_offset) + ib;
+
+    dst_t * y = yy + ib * QK_K + 128 * ip + il;
+
+    const uint8_t * ql = ql_ptr + 64 * ip + il;
+    const uint8_t   qh = *(qh_ptr + 32 * ip + il);
+    const int8_t *  sc = reinterpret_cast<const int8_t *>(scales_ptr + is);
+
+    y[0]  = *d * sc[0] * ((int8_t) ((ql[0] & 0xF) | (((qh >> 0) & 3) << 4)) - 32);
+    y[32] = *d * sc[2] * ((int8_t) ((ql[32] & 0xF) | (((qh >> 2) & 3) << 4)) - 32);
+    y[64] = *d * sc[4] * ((int8_t) ((ql[0] >> 4) | (((qh >> 4) & 3) << 4)) - 32);
+    y[96] = *d * sc[6] * ((int8_t) ((ql[32] >> 4) | (((qh >> 6) & 3) << 4)) - 32);
+}
+
 template<typename dst_t>
 static void dequantize_block_iq2_xxs(const void * __restrict__ vx, dst_t * __restrict__ yy,
                                      const sycl::nd_item<3> &item_ct1,

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -354,7 +354,8 @@ ggml_backend_sycl_buffer_init_tensor(ggml_backend_buffer_t buffer,
         assert(tensor->view_src->buffer->buft == buffer->buft);
         return GGML_STATUS_SUCCESS;
     }
-    if ((tensor->type == GGML_TYPE_Q4_0 || tensor->type == GGML_TYPE_Q4_K) && !g_ggml_sycl_disable_optimize) {
+    if ((tensor->type == GGML_TYPE_Q4_0 || tensor->type == GGML_TYPE_Q4_K || tensor->type == GGML_TYPE_Q6_K) &&
+        !g_ggml_sycl_disable_optimize) {
         ggml_tensor_extra_gpu * extra = new ggml_tensor_extra_gpu{};
         tensor->extra                 = extra;
         ctx->tensor_extras.push_back(extra);  //used to release it when destroy ctx.
@@ -2989,6 +2990,7 @@ inline bool ggml_sycl_supports_reorder_mul_mat_sycl(enum ggml_type type) {
         case GGML_TYPE_Q4_0:
             return true;
         case GGML_TYPE_Q4_K:
+        case GGML_TYPE_Q6_K:
             return !g_ggml_sycl_prioritize_dmmv;
         default:
             return false;
@@ -3008,6 +3010,7 @@ inline bool ggml_sycl_supports_reorder_mmvq(enum ggml_type type) {
     switch (type) {
         case GGML_TYPE_Q4_0:
         case GGML_TYPE_Q4_K:
+        case GGML_TYPE_Q6_K:
             return true;
         default:
             return false;
@@ -3092,6 +3095,50 @@ static void reorder_qw_q4_k(uint8_t * data_device, size_t size, size_t offset, d
     sycl::free(tmp_buf, *stream);
 }
 
+static void reorder_qw_q6_k(uint8_t * data_device, size_t size, size_t offset, dpct::queue_ptr stream) {
+    GGML_ASSERT(size % sizeof(block_q6_K) == 0);
+    GGML_ASSERT(offset % sizeof(block_q6_K) == 0);
+
+    const int nblocks = size / sizeof(block_q6_K);
+
+    auto * tmp_buf = sycl::malloc_shared<uint8_t>(size, *stream);
+    SYCL_CHECK(CHECK_TRY_ERROR((*stream).memcpy(tmp_buf, data_device, size).wait()));
+
+    auto *       ql_ptr     = data_device;
+    auto *       qh_ptr     = ql_ptr + (QK_K / 2) * nblocks;
+    auto *       scales_ptr = qh_ptr + (QK_K / 4) * nblocks;
+    sycl::half * dm_ptr     = (sycl::half *) (scales_ptr + (QK_K / 16) * nblocks);
+
+    stream
+        ->parallel_for(nblocks,
+                       [=](auto i) {
+                           const block_q6_K * x  = (const block_q6_K *) tmp_buf;
+                           const int          ib = i;
+
+                           const uint8_t * ql              = x[ib].ql;
+                           const uint8_t * qh              = x[ib].qh;
+                           uint8_t *       base_ql_ptr     = ql_ptr + (QK_K / 2) * ib;
+                           uint8_t *       base_qh_ptr     = qh_ptr + (QK_K / 4) * ib;
+                           uint8_t *       base_scales_ptr = scales_ptr + (QK_K / 16) * ib;
+
+                           for (int j = 0; j < QK_K / 2; ++j) {
+                               base_ql_ptr[j] = ql[j];
+                           }
+                           for (int j = 0; j < QK_K / 4; ++j) {
+                               base_qh_ptr[j] = qh[j];
+                           }
+
+                           for (int j = 0; j < QK_K / 16; ++j) {
+                               base_scales_ptr[j] = x[ib].scales[j];
+                           }
+
+                           dm_ptr[ib] = x[ib].d;
+                       })
+        .wait_and_throw();
+
+    sycl::free(tmp_buf, *stream);
+}
+
 static void reorder_qw(const ggml_tensor * src0, dpct::queue_ptr stream) {
     uint8_t * data_device = (uint8_t *) src0->data;
     size_t ncols = src0->ne[0];
@@ -3105,6 +3152,9 @@ static void reorder_qw(const ggml_tensor * src0, dpct::queue_ptr stream) {
         case GGML_TYPE_Q4_K:
             reorder_qw_q4_k(data_device, size, 0, stream);
             break;
+        case GGML_TYPE_Q6_K:
+            reorder_qw_q6_k(data_device, size, 0, stream);
+            break;
         default:
             GGML_ABORT("reorder_qw() called with unsupported type");
             break;
@@ -4226,6 +4276,9 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
             {
                 ggml_type src0_type = op->src[0]->type;
                 ggml_type src1_type = op->src[1]->type;
+                if (src0_type == src1_type && (ggml_is_contiguous(op->src[0]) && ggml_is_contiguous(op->src[1])) && src0_type != GGML_TYPE_BF16) {
+                    return true;
+                }
                 if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_F32) {
                     return true;
                 }
@@ -4271,6 +4324,21 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
                 if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_IQ4_NL) {
                     return true;
                 }
+                if(src0_type == GGML_TYPE_Q8_0 && src1_type == GGML_TYPE_Q8_0) {
+                    return true;
+                }
+                if(src0_type == GGML_TYPE_Q5_0 && src1_type == GGML_TYPE_Q5_0) {
+                    return true;
+                }
+                if(src0_type == GGML_TYPE_Q5_1 && src1_type == GGML_TYPE_Q5_1) {
+                    return true;
+                }
+                if(src0_type == GGML_TYPE_Q4_0 && src1_type == GGML_TYPE_Q4_0) {
+                    return true;
+                }
+                if(src0_type == GGML_TYPE_Q4_1 && src1_type == GGML_TYPE_Q4_1) {
+                    return true;
+                }
                 return false;
             }
         case GGML_OP_CONCAT:

ggml/src/ggml-sycl/mmvq.cpp

@@ -31,11 +31,10 @@ static void mul_mat_vec_q_reorder(const void * __restrict__ vx, const void * __r
 
     float partial_sum = 0.0f;
     for (int i = sg.get_local_linear_id() / block_elements_per_subgroup; i < blocks_per_row; i += blocks_per_subgroup) {
-        const int ibx       = row * blocks_per_row + i;  // x block index
-        // TODO: Generalize offsets, right now only works for quantizations that don't split high and low bits
-        const int bx_offset = block_type::get_block_offset(ibx);
-        const int d_offset  = block_type::get_d_offset(nrows, ncols, ibx);
+        const int ibx = row * blocks_per_row + i;  // x block index
 
+        const auto         bx_offset      = block_type::get_block_offset(ibx, nblocks);
+        const auto         d_offset       = block_type::get_d_offset(nrows, ncols, ibx);
         // Y block index that aligns with ibx
         const int iby = i * block_type::block_to_q8_1_ratio();
         const int8_t* q8_1_quant_ptr = (const int8_t*)vy + iby * QK8_1;
@@ -46,7 +45,7 @@ static void mul_mat_vec_q_reorder(const void * __restrict__ vx, const void * __r
             // x block quant index when casting the quants to int
             const int iqs = elem + block_traits::vdr_mmvq * (sg.get_local_linear_id() % block_elements_per_subgroup);
 
-            partial_sum += reorder_vec_dot_q_sycl()(vx, bx_offset, d_offset, q8_1_quant_ptr, q8_1_ds_ptr, iqs, nblocks);
+            partial_sum += reorder_vec_dot_q_sycl()(vx, bx_offset, d_offset, q8_1_quant_ptr, q8_1_ds_ptr, iqs);
         }
     }
 
@@ -785,6 +784,24 @@ static void mul_mat_vec_q5_K_q8_1_sycl(const void *vx, const void *vy,
     }
 }
 
+static void reorder_mul_mat_vec_q6_k_q8_1_sycl(const void * vx, const void * vy, float * dst, const int ncols,
+                                               const int nrows, dpct::queue_ptr stream) {
+    GGML_ASSERT(ncols % QK_K == 0);
+    const int        block_num_y   = ceil_div(nrows, GGML_SYCL_MMV_Y);
+    constexpr size_t num_subgroups = 16;
+    GGML_ASSERT(block_num_y % num_subgroups == 0);
+
+    const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, block_num_y * WARP_SIZE);
+    const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE);
+
+    stream->submit([&](sycl::handler & cgh) {
+        cgh.parallel_for(sycl::nd_range<3>(global_size, workgroup_size),
+                         [=](sycl::nd_item<3> nd_item) [[sycl::reqd_sub_group_size(WARP_SIZE)]] {
+                             mul_mat_vec_q_reorder<reorder_vec_dot_q_sycl<GGML_TYPE_Q6_K>>(vx, vy, dst, ncols, nrows,
+                                                                                           nd_item);
+                         });
+    });
+}
 static void mul_mat_vec_q6_K_q8_1_sycl(const void *vx, const void *vy,
                                        float *dst, const int ncols,
                                        const int nrows,
@@ -1070,7 +1087,14 @@ void ggml_sycl_op_mul_mat_vec_q(ggml_backend_sycl_context & ctx, const ggml_tens
                 mul_mat_vec_q5_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
                 break;
             case GGML_TYPE_Q6_K:
-                mul_mat_vec_q6_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+                if ((ggml_tensor_extra_gpu *) dst->src[0]->extra &&
+                    ((ggml_tensor_extra_gpu *) dst->src[0]->extra)->optimized_feature.reorder) {
+                    GGML_SYCL_DEBUG("Calling reorder_mul_mat_vec_q6_k_q8_1_sycl\n");
+                    reorder_mul_mat_vec_q6_k_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+                } else {
+                    GGML_SYCL_DEBUG("Calling mul_mat_vec_q6_k_q8_1_sycl\n");
+                    mul_mat_vec_q6_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+                }
                 break;
             case GGML_TYPE_IQ1_S:
                 mul_mat_vec_iq1_s_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);

ggml/src/ggml-sycl/quants.hpp

@@ -14,12 +14,13 @@
 #ifndef GGML_SYCL_QUANTS_HPP
 #define GGML_SYCL_QUANTS_HPP
 
+#include <utility>
+
 #include "ggml-common.h"
 #include "ggml.h"
 
 namespace ggml_sycl_reordered {
 
-
 // The reordered block moves quants (qs) and  scales(d) to two
 // uniform regions of memory that is contiguous in the same tensor.
 // What this means is that instead of having:
@@ -32,7 +33,6 @@ namespace ggml_sycl_reordered {
 
 template <ggml_type type> struct block_q_t;
 
-
 // qk number of weights / quants in a block
 // qr number of weights in a byte (described as 'before dequantization')
 //    for quantization types that has low and high bits split, qr is calculated with
@@ -47,10 +47,12 @@ template <> struct block_q_t<GGML_TYPE_Q4_0> {
         static constexpr uint32_t vdr_mmvq = 2;
     };
 
-    static constexpr int get_block_offset(const int block_index) { return block_index * (traits::qk / traits::qr); }
+    static constexpr std::pair<int, int> get_block_offset(const int block_index, const int /* nblocks */) {
+        return { block_index * (traits::qk / traits::qr), 0 };
+    }
 
-    static constexpr int get_d_offset(int nrows, int ncols, const int block_index) {
-        return (ncols / traits::qr * nrows) + block_index * sizeof(ggml_half);
+    static constexpr std::pair<int, int> get_d_offset(int nrows, int ncols, const int block_index) {
+        return { (ncols / traits::qr * nrows) + block_index * sizeof(ggml_half), 0 };
     }
 
     static constexpr int block_to_q8_1_ratio() { return traits::qk / QK8_1; }
@@ -64,20 +66,46 @@ template <> struct block_q_t<GGML_TYPE_Q4_K> {
         static constexpr uint32_t vdr_mmvq = 2;
     };
 
-    static constexpr int get_block_offset(const int block_index) { return block_index * (traits::qk / traits::qr); }
+    static constexpr std::pair<int, int> get_block_offset(const int block_index, const int /* nblocks */) {
+        return { block_index * (traits::qk / traits::qr), 0 };
+    }
 
-    static constexpr int get_d_offset(int nrows, int ncols, const int block_index) {
+    static constexpr std::pair<int, int> get_d_offset(int nrows, int ncols, const int block_index) {
         auto nblocks = (nrows * (ncols / traits::qk));
-        return (nblocks * QK_K / 2) + (nblocks * K_SCALE_SIZE) + (block_index * sizeof(ggml_half2));
+        return { nblocks * (QK_K / 2),
+                 (nblocks * QK_K / 2) + (nblocks * K_SCALE_SIZE) + (block_index * sizeof(ggml_half2)) };
     }
 
     static constexpr int block_to_q8_1_ratio() { return traits::qk / QK8_1; }
 
     constexpr size_t get_total_qs_bytes(int nblocks) { return nblocks * QK_K / 2; }
-
-    constexpr size_t get_dm_offset(int nblocks) { return get_total_qs_bytes(nblocks) + nblocks * K_SCALE_SIZE; }
 };
 
+template <> struct block_q_t<GGML_TYPE_Q6_K> {
+    struct traits {
+        static constexpr uint32_t qk       = QK_K;
+        static constexpr uint32_t qi       = QI6_K;
+        static constexpr uint32_t qr       = QR6_K;
+        static constexpr uint32_t vdr_mmvq = 1;
+    };
+
+    static constexpr std::pair<int, int> get_block_offset(const int block_index, const int n_blocks) {
+        auto low_bits_index  = block_index * (traits::qk / traits::qr);
+        // the index of high bits it's after all low bits
+        auto high_bits_index = n_blocks * (QK_K / 2) + (block_index * (QK_K / 4));
+        return { low_bits_index, high_bits_index };
+    }
+
+    static constexpr std::pair<int, int> get_d_offset(int nrows, int ncols, const int block_index) {
+        auto nblocks        = (nrows * (ncols / traits::qk));
+        auto total_qs_bytes = nblocks * (QK_K / 2) + nblocks * (QK_K / 4);
+        auto block_scales   = total_qs_bytes + block_index * (QK_K / 16);
+        auto sb_scale       = total_qs_bytes + nblocks * (QK_K / 16);
+        return { block_scales, sb_scale };
+    }
+
+    static constexpr int block_to_q8_1_ratio() { return traits::qk / QK8_1; }
+};
 }  // namespace ggml_sycl_reordered
 
 #endif  // GGML_SYCL_QUANTS_HPP

ggml/src/ggml-sycl/vecdotq.hpp

@@ -284,10 +284,11 @@ template <> struct reorder_vec_dot_q_sycl<GGML_TYPE_Q4_0> {
         return d4 * (sumi * ds8f.x() - (8 * q4_0_traits::vdr_mmvq / q4_0_traits::qi) * ds8f.y());
     }
 
-    __dpct_inline__ float operator()(const void * __restrict__ vbq, const int ibx_offset, const int d_offset,
-                     const int8_t* q8_1_quant_ptr, const sycl::half2* q8_1_ds, const int & iqs, int /* nblocks */) {
-        const uint8_t * bq4_0 = static_cast<const uint8_t *>(vbq) + ibx_offset;
-        const ggml_half d     = *(reinterpret_cast<const ggml_half *>(static_cast<const uint8_t *>(vbq) + d_offset));
+    __dpct_inline__ float operator()(const void * __restrict__ vbq, const std::pair<int, int> ibx_offset,
+                                     const std::pair<int, int> d_offset, const int8_t * q8_1_quant_ptr,
+                                     const sycl::half2 * q8_1_ds, const int & iqs) {
+        const uint8_t * bq4_0 = static_cast<const uint8_t *>(vbq) + ibx_offset.first;
+        const ggml_half d = *(reinterpret_cast<const ggml_half *>(static_cast<const uint8_t *>(vbq) + d_offset.first));
         int             v[q4_0_traits::vdr_mmvq];
         int             u[2 * q4_0_traits::vdr_mmvq];
 
@@ -346,15 +347,15 @@ template <> struct reorder_vec_dot_q_sycl<GGML_TYPE_Q4_K> {
     using q4_k_block  = ggml_sycl_reordered::block_q_t<GGML_TYPE_Q4_K>;
     using q4_k_traits = typename q4_k_block::traits;
 
-    float operator()(const void * __restrict__ vbq, const int ibx_offset, const int d_offset,
-                     const int8_t* q8_1_quant_ptr, const sycl::half2* q8_1_ds, const int & iqs, int nblocks) {
-        const int ib = ibx_offset / (QK_K / 2);
+    __dpct_inline__ float operator()(const void * __restrict__ vbq, const std::pair<int, int> ibx_offset,
+                                     const std::pair<int, int> d_offset, const int8_t * q8_1_quant_ptr,
+                                     const sycl::half2 * q8_1_ds, const int & iqs) {
+        const int ib = ibx_offset.first / (QK_K / 2);
 
         const uint8_t *    base           = static_cast<const uint8_t *>(vbq);
-        const uint8_t *    qs             = base + ibx_offset;
-        const int          total_qs_bytes = nblocks * (QK_K / 2);
-        const uint8_t *    scs            = base + total_qs_bytes + ib * K_SCALE_SIZE;
-        const ggml_half2 * dms            = reinterpret_cast<const ggml_half2 *>(base + d_offset);
+        const uint8_t *    qs             = base + ibx_offset.first;
+        const uint8_t *    scs            = base + d_offset.first + ib * K_SCALE_SIZE;
+        const ggml_half2 * dms            = reinterpret_cast<const ggml_half2 *>(base + d_offset.second);
 
         const int        bq8_offset = QR4_K * ((iqs / 2) / (QI8_1 / 2));
         const int *      q4         = (const int *) (qs + 16 * bq8_offset + 4 * ((iqs / 2) % 4));
@@ -395,6 +396,66 @@ template <> struct reorder_vec_dot_q_sycl<GGML_TYPE_Q4_K> {
     }
 };
 
+template <> struct reorder_vec_dot_q_sycl<GGML_TYPE_Q6_K> {
+    static constexpr ggml_type gtype = GGML_TYPE_Q6_K;
+
+    using q6_k_block  = ggml_sycl_reordered::block_q_t<GGML_TYPE_Q6_K>;
+    using q6_k_traits = typename q6_k_block::traits;
+
+    __dpct_inline__ float vec_dot_q6_K_q8_1_impl_mmvq(const int vl, const int vh, const int * __restrict__ u,
+                                                      const int8_t * __restrict__ scales, const float d,
+                                                      const float * __restrict__ d8) {
+        float sumf = 0.0f;
+
+#pragma unroll
+        for (int i = 0; i < QR6_K; ++i) {
+            const int sc = scales[4 * i];
+
+            const int vil = (vl >> (4 * i)) & 0x0F0F0F0F;
+
+            const int vih = ((vh >> (4 * i)) << 4) & 0x30303030;
+
+            const int vi = dpct::vectorized_binary<sycl::char4>((vil | vih), 0x20202020,
+                                                                dpct::sub_sat());  // vi = (vil | vih) - 32
+
+            sumf += d8[i] * (dpct::dp4a(vi, u[i], 0) * sc);                        // SIMD dot product
+        }
+
+        return d * sumf;
+    }
+
+    __dpct_inline__ float operator()(const void * __restrict__ vbq, const std::pair<int, int> ibx_offset,
+                     const std::pair<int, int> d_offset, const int8_t * q8_1_quant_ptr, const sycl::half2 * q8_1_ds,
+                     const int iqs) {
+        const int ib = ibx_offset.first / (QK_K / 2);
+
+        const uint8_t *   base   = static_cast<const uint8_t *>(vbq);
+        const uint8_t *   ql     = base + ibx_offset.first;
+        const uint8_t *   qh     = base + ibx_offset.second;
+        const int8_t *    scales = reinterpret_cast<const int8_t *>(base + d_offset.first);
+        const ggml_half * d      = (const ggml_half *) (base + d_offset.second) + ib;
+
+        const int bq8_offset   = 2 * QR6_K * (iqs / (QI6_K / 2)) + (iqs % (QI6_K / 2)) / (QI6_K / 4);
+        const int scale_offset = (QI6_K / 4) * (iqs / (QI6_K / 2)) + (iqs % (QI6_K / 2)) / (QI6_K / 8);
+        const int vh_shift     = 2 * ((iqs % (QI6_K / 2)) / (QI6_K / 4));
+
+        const int vl = get_int_from_uint8(ql, iqs);
+        const int vh = get_int_from_uint8(qh, (QI6_K / 4) * (iqs / (QI6_K / 2)) + iqs % (QI6_K / 4)) >> vh_shift;
+
+        const int8_t * scs = scales + scale_offset;
+
+        int   u[QR6_K];
+        float d8[QR6_K];
+
+#pragma unroll
+        for (int i = 0; i < QR6_K; ++i) {
+            u[i] = get_int_from_int8_aligned(q8_1_quant_ptr + (bq8_offset + 2 * i) * QK8_1, iqs % QI8_1);
+            const sycl::half2 ds_values = *(q8_1_ds + bq8_offset + 2 * i);
+            d8[i]                       = ds_values[0];
+        }
+        return vec_dot_q6_K_q8_1_impl_mmvq(vl, vh, u, scs, *d, d8);
+    }
+};
 #define VDR_Q4_0_Q8_1_MMVQ 2
 #define VDR_Q4_0_Q8_1_MMQ  4
 

ggml/src/ggml-vulkan/ggml-vulkan.cpp

@@ -196,6 +196,7 @@ enum vk_device_architecture {
     AMD_RDNA1,
     AMD_RDNA2,
     AMD_RDNA3,
+    INTEL_XE2,
 };
 
 static vk_device_architecture get_device_architecture(const vk::PhysicalDevice& device) {
@@ -246,6 +247,34 @@ static vk_device_architecture get_device_architecture(const vk::PhysicalDevice&
             }
             return vk_device_architecture::AMD_RDNA2;
         }
+    } else if (props.vendorID == VK_VENDOR_ID_INTEL) {
+        const std::vector<vk::ExtensionProperties> ext_props = device.enumerateDeviceExtensionProperties();
+
+        bool subgroup_size_control = false;
+
+        for (const auto& properties : ext_props) {
+            if (strcmp("VK_EXT_subgroup_size_control", properties.extensionName) == 0) {
+                subgroup_size_control = true;
+            }
+        }
+
+        if (!subgroup_size_control) {
+            return vk_device_architecture::OTHER;
+        }
+
+        vk::PhysicalDeviceProperties2 props2;
+        vk::PhysicalDeviceSubgroupSizeControlPropertiesEXT subgroup_size_control_props;
+
+        props2.pNext = &subgroup_size_control_props;
+        device.getProperties2(&props2);
+
+        if (subgroup_size_control_props.minSubgroupSize == 16) {
+            // Xe2 architecture uses SIMD16 while previous Xe and Gen architecture uses SIMD8.
+            // Minimum subgroup size matches the SIMD width so we distinguish architecture by checking this value.
+            // https://www.intel.com/content/www/us/en/content-details/824434/2024-intel-tech-tour-xe2-and-lunar-lake-s-gpu.html
+            // https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2025-0/intel-xe-gpu-architecture.html
+            return vk_device_architecture::INTEL_XE2;
+        }
     }
     return vk_device_architecture::OTHER;
 }
@@ -396,6 +425,7 @@ struct vk_device_struct {
     vk_pipeline pipeline_count_equal_i32;
     vk_pipeline pipeline_im2col_f32, pipeline_im2col_f32_f16;
     vk_pipeline pipeline_timestep_embedding_f32;
+    vk_pipeline pipeline_conv_transpose_1d_f32;
     vk_pipeline pipeline_pool2d_f32;
     vk_pipeline pipeline_rwkv_wkv6_f32;
     vk_pipeline pipeline_rwkv_wkv7_f32;
@@ -444,7 +474,7 @@ struct vk_device_struct {
     // for GGML_VK_PERF_LOGGER
     std::unique_ptr<vk_perf_logger> perf_logger;
     vk::QueryPool query_pool;
-    uint32_t num_queries;
+    int32_t num_queries;
 
     ~vk_device_struct() {
         VK_LOG_DEBUG("destroy device " << name);
@@ -706,6 +736,21 @@ struct vk_op_timestep_embedding_push_constants {
     uint32_t max_period;
 };
 
+struct vk_op_conv_transpose_1d_push_constants {
+    uint32_t Cout;
+    uint32_t Cin;
+    uint32_t K;
+    uint32_t L;
+    uint32_t KL;
+
+    uint32_t nb01;
+    uint32_t nb02;
+    uint32_t nb11;
+    uint32_t nb1;
+
+    int32_t s0;
+};
+
 struct vk_op_pool2d_push_constants {
     uint32_t IW; uint32_t IH;
     uint32_t OW; uint32_t OH;
@@ -2726,6 +2771,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
 
     ggml_vk_create_pipeline(device, device->pipeline_timestep_embedding_f32, "timestep_embedding_f32", timestep_embedding_f32_len, timestep_embedding_f32_data, "main", 2, sizeof(vk_op_timestep_embedding_push_constants), {256, 1, 1}, {}, 1);
 
+    ggml_vk_create_pipeline(device, device->pipeline_conv_transpose_1d_f32, "conv_transpose_1d_f32", conv_transpose_1d_f32_len, conv_transpose_1d_f32_data, "main", 3, sizeof(vk_op_conv_transpose_1d_push_constants), {1, 1, 1}, {}, 1);
+
     ggml_vk_create_pipeline(device, device->pipeline_pool2d_f32, "pool2d_f32", pool2d_f32_len, pool2d_f32_data, "main", 2, sizeof(vk_op_pool2d_push_constants), {512, 1, 1}, {}, 1);
 
     ggml_vk_create_pipeline(device, device->pipeline_rwkv_wkv6_f32, "rwkv_wkv6_f32", rwkv_wkv6_f32_len, rwkv_wkv6_f32_data, "main", 7, sizeof(vk_op_rwkv_wkv6_push_constants), {1, 1, 1}, {device->subgroup_size}, 1);
@@ -4061,7 +4108,33 @@ static vk_submission ggml_vk_begin_submission(vk_device& device, vk_queue& q, bo
     return s;
 }
 
-static void ggml_vk_dispatch_pipeline(ggml_backend_vk_context* ctx, vk_context& subctx, vk_pipeline& pipeline, std::initializer_list<vk::DescriptorBufferInfo> const& descriptor_buffer_infos, size_t push_constant_size, const void* push_constants, std::array<uint32_t, 3> elements) {
+template <typename T> size_t push_constant_size(const T &t) {
+    static_assert(std::is_class<T>::value, "T must be a struct/class");
+    GGML_UNUSED(t);
+    return sizeof(T);
+}
+template <typename T> size_t push_constant_size(const std::vector<T> &t) {
+    GGML_UNUSED(t);
+    return sizeof(T) * t.size();
+}
+template <typename T, uint32_t N> size_t push_constant_size(const std::array<T, N> &t) {
+    GGML_UNUSED(t);
+    return sizeof(T) * N;
+}
+
+template <typename T> const T *push_constant_data(const T &t) {
+    static_assert(std::is_class<T>::value, "T must be a struct/class");
+    return &t;
+}
+template <typename T> const T *push_constant_data(const std::vector<T> &t) {
+    return t.data();
+}
+template <typename T, uint32_t N> const T *push_constant_data(const std::array<T, N> &t) {
+    return t.data();
+}
+
+template <typename T>
+static void ggml_vk_dispatch_pipeline(ggml_backend_vk_context* ctx, vk_context& subctx, vk_pipeline& pipeline, std::initializer_list<vk::DescriptorBufferInfo> const& descriptor_buffer_infos, const T &push_constants, std::array<uint32_t, 3> elements) {
     const uint32_t wg0 = CEIL_DIV(elements[0], pipeline->wg_denoms[0]);
     const uint32_t wg1 = CEIL_DIV(elements[1], pipeline->wg_denoms[1]);
     const uint32_t wg2 = CEIL_DIV(elements[2], pipeline->wg_denoms[2]);
@@ -4077,7 +4150,7 @@ static void ggml_vk_dispatch_pipeline(ggml_backend_vk_context* ctx, vk_context&
     vk::WriteDescriptorSet write_descriptor_set{ descriptor_set, 0, 0, pipeline->parameter_count, vk::DescriptorType::eStorageBuffer, nullptr, descriptor_buffer_infos.begin() };
     ctx->device->device.updateDescriptorSets({ write_descriptor_set }, {});
 
-    subctx->s->buffer.pushConstants(pipeline->layout, vk::ShaderStageFlagBits::eCompute, 0, push_constant_size, push_constants);
+    subctx->s->buffer.pushConstants(pipeline->layout, vk::ShaderStageFlagBits::eCompute, 0, push_constant_size(push_constants), push_constant_data(push_constants));
     subctx->s->buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline->pipeline);
     subctx->s->buffer.bindDescriptorSets(vk::PipelineBindPoint::eCompute,
                                 pipeline->layout,
@@ -4540,7 +4613,7 @@ static void ggml_vk_matmul(
     ggml_vk_sync_buffers(subctx);
     if (split_k == 1) {
         const vk_mat_mat_push_constants pc = { m, n, k, stride_a, stride_b, stride_d, batch_stride_a, batch_stride_b, batch_stride_d, k, ne02, ne12, broadcast2, broadcast3, padded_n };
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, d }, sizeof(vk_mat_mat_push_constants), &pc, { m, n, batch });
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, d }, pc, { m, n, batch });
         return;
     }
 
@@ -4548,10 +4621,10 @@ static void ggml_vk_matmul(
 
     const vk_mat_mat_push_constants pc1 = { m, n, k, stride_a, stride_b, stride_d, batch_stride_a, batch_stride_b, batch_stride_d, CEIL_DIV(k, split_k), ne02, ne12, broadcast2, broadcast3, padded_n };
     // Make sure enough workgroups get assigned for split k to work
-    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, split_k_buffer }, sizeof(vk_mat_mat_push_constants), &pc1, { (CEIL_DIV(m, pipeline->wg_denoms[0]) * pipeline->wg_denoms[0]) * split_k, n, batch });
+    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, split_k_buffer }, pc1, { (CEIL_DIV(m, pipeline->wg_denoms[0]) * pipeline->wg_denoms[0]) * split_k, n, batch });
     ggml_vk_sync_buffers(subctx);
     const std::array<uint32_t, 2> pc2 = { (uint32_t)(m * n * batch), split_k };
-    ggml_vk_dispatch_pipeline(ctx, subctx, ctx->device->pipeline_matmul_split_k_reduce, { split_k_buffer, d }, pc2.size() * sizeof(uint32_t), pc2.data(), { m * n * batch, 1, 1 });
+    ggml_vk_dispatch_pipeline(ctx, subctx, ctx->device->pipeline_matmul_split_k_reduce, { split_k_buffer, d }, pc2, { m * n * batch, 1, 1 });
 }
 
 static vk_pipeline ggml_vk_guess_matmul_id_pipeline(ggml_backend_vk_context * ctx, vk_matmul_pipeline& mmp, uint32_t m, uint32_t n, bool aligned, ggml_type src0_type) {
@@ -4599,7 +4672,7 @@ static void ggml_vk_matmul_id(
     ggml_vk_sync_buffers(subctx);
     const vk_mat_mat_id_push_constants pc = { m, n, k, stride_a, stride_b, stride_d, batch_stride_a, batch_stride_b, batch_stride_d,
                                               nei0, nei1, nbi1, ne11, padded_n };
-    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, d, ids }, sizeof(vk_mat_mat_id_push_constants), &pc, { m, nei1, n_as });
+    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { a, b, d, ids }, pc, { m, nei1, n_as });
 }
 
 static bool ggml_vk_dim01_contiguous(const ggml_tensor * tensor) {
@@ -4720,7 +4793,7 @@ static void ggml_vk_cpy_to_contiguous(ggml_backend_vk_context * ctx, vk_context&
     };
     init_pushconst_fastdiv(pc);
     ggml_vk_sync_buffers(subctx);
-    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { in, out }, sizeof(vk_op_unary_push_constants), &pc, elements);
+    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { in, out }, pc, elements);
 }
 
 static vk_pipeline ggml_vk_get_quantize_pipeline(ggml_backend_vk_context * ctx, ggml_type type) {
@@ -4739,7 +4812,7 @@ static void ggml_vk_quantize_q8_1(ggml_backend_vk_context * ctx, vk_context& sub
     vk_pipeline pipeline = ggml_vk_get_quantize_pipeline(ctx, GGML_TYPE_Q8_1);
 
     ggml_vk_sync_buffers(subctx);
-    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { in, out }, sizeof(uint32_t), &ne, { ne, 1, 1 });
+    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { in, out }, std::array<uint32_t, 1>{ne}, { ne, 1, 1 });
 }
 
 static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
@@ -4939,7 +5012,7 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
     } else if (qx_needs_dequant) {
         const std::vector<uint32_t> pc = { (uint32_t)ne01, (uint32_t)ne10, (uint32_t)ne10, (uint32_t)ne10, (uint32_t)(ggml_nelements(src0)) };
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, to_fp16_vk_0, { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz * ne02 * ne03 }, vk_subbuffer{ d_X, 0, x_sz * ne02 * ne03 } }, pc.size() * sizeof(uint32_t), pc.data(), { (uint32_t)(x_ne * ne02 * ne03), 1, 1});
+        ggml_vk_dispatch_pipeline(ctx, subctx, to_fp16_vk_0, { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz * ne02 * ne03 }, vk_subbuffer{ d_X, 0, x_sz * ne02 * ne03 } }, pc, { (uint32_t)(x_ne * ne02 * ne03), 1, 1});
     }
     if (y_non_contig) {
         ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
@@ -5155,7 +5228,7 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
     ggml_vk_sync_buffers(subctx);
     ggml_vk_dispatch_pipeline(ctx, subctx, dmmv,
                               { vk_subbuffer{ d_X, x_buf_offset, x_sz * ne02 * ne03 }, vk_subbuffer{ d_Y, y_buf_offset, y_sz * ne12 * ne13 }, vk_subbuffer{ d_D, d_buf_offset, d_sz * ne22 * ne23} },
-                              sizeof(vk_mat_vec_push_constants), &pc, { groups_x, (uint32_t)(ne12 * ne13), groups_z });
+                              pc, { groups_x, (uint32_t)(ne12 * ne13), groups_z });
 }
 
 static void ggml_vk_mul_mat_vec_p021_f16_f32(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
@@ -5243,7 +5316,7 @@ static void ggml_vk_mul_mat_vec_p021_f16_f32(ggml_backend_vk_context * ctx, vk_c
     }
 
     ggml_vk_sync_buffers(subctx);
-    ggml_vk_dispatch_pipeline(ctx, subctx, ctx->device->pipeline_mul_mat_vec_p021_f16_f32[gqa_ratio - 1], { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz }, vk_subbuffer{ d_Qy, qy_buffer_offset, qy_sz + qy_shader_offset }, vk_subbuffer{ d_D, d_buffer_offset, d_sz + d_shader_offset } }, 6 * sizeof(uint32_t), &pc, { 1, (uint32_t)ne01, workgroups_z });
+    ggml_vk_dispatch_pipeline(ctx, subctx, ctx->device->pipeline_mul_mat_vec_p021_f16_f32[gqa_ratio - 1], { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz }, vk_subbuffer{ d_Qy, qy_buffer_offset, qy_sz + qy_shader_offset }, vk_subbuffer{ d_D, d_buffer_offset, d_sz + d_shader_offset } }, pc, { 1, (uint32_t)ne01, workgroups_z });
 }
 
 static void ggml_vk_mul_mat_vec_nc_f16_f32(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
@@ -5326,7 +5399,7 @@ static void ggml_vk_mul_mat_vec_nc_f16_f32(ggml_backend_vk_context * ctx, vk_con
     const std::array<uint32_t, 9> pc = { (uint32_t)ne00, (uint32_t)ne01, row_stride_x, channel_stride_x, channel_stride_y, (uint32_t)(ne12 / ne02), (uint32_t)ne12, (uint32_t)(qy_shader_offset / ggml_type_size(src1->type)), (uint32_t)(d_shader_offset / ggml_type_size(dst->type)) };
     ggml_vk_sync_buffers(subctx);
     ggml_vk_dispatch_pipeline(ctx, subctx, ctx->device->pipeline_mul_mat_vec_nc_f16_f32,
-        { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz }, vk_subbuffer{ d_Qy, qy_buffer_offset, qy_sz + qy_shader_offset }, vk_subbuffer{ d_D, d_buffer_offset, d_sz + d_shader_offset } }, 7 * sizeof(uint32_t), &pc, { 1, (uint32_t)ne01, (uint32_t)ne12 });
+        { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz }, vk_subbuffer{ d_Qy, qy_buffer_offset, qy_sz + qy_shader_offset }, vk_subbuffer{ d_D, d_buffer_offset, d_sz + d_shader_offset } }, pc, { 1, (uint32_t)ne01, (uint32_t)ne12 });
 }
 
 static void ggml_vk_mul_mat(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
@@ -5542,7 +5615,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
         const std::vector<uint32_t> pc = { (uint32_t)ne01, (uint32_t)ne10, (uint32_t)ne10, (uint32_t)ne10, (uint32_t)(ggml_nelements(src0)) };
         ggml_vk_sync_buffers(subctx);
         ggml_vk_dispatch_pipeline(ctx, subctx, to_fp16_vk_0,
-            { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz * ne02 * ne03 }, vk_subbuffer{ d_X, 0, x_sz * ne02 * ne03 } }, pc.size() * sizeof(uint32_t), pc.data(), { (uint32_t)(x_ne * ne02 * ne03), 1, 1});
+            { vk_subbuffer{ d_Qx, qx_buf_offset, qx_sz * ne02 * ne03 }, vk_subbuffer{ d_X, 0, x_sz * ne02 * ne03 } }, pc, { (uint32_t)(x_ne * ne02 * ne03), 1, 1});
     }
     if (y_non_contig) {
         ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
@@ -5762,7 +5835,7 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte
     ggml_vk_dispatch_pipeline(ctx, subctx, dmmv,
         { vk_subbuffer{ d_X, x_buf_offset, x_sz * ne02 * ne03 },
         vk_subbuffer{ d_Y, y_buf_offset, y_sz * ne12 * ne13 }, vk_subbuffer{ d_D, d_buf_offset, d_sz * ne22 * ne23}, vk_subbuffer{ d_ids, ids_buf_offset, ids_sz } },
-        sizeof(vk_mat_vec_id_push_constants), &pc, { groups_x, (uint32_t)nei0, groups_z });
+        pc, { groups_x, (uint32_t)nei0, groups_z });
 }
 
 static void ggml_vk_mul_mat_id(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst, bool dryrun = false) {
@@ -6112,7 +6185,7 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
                                     // there's no more than one tile of rows (i.e. workgroups_x would have been
                                     // one). We reuse workgroups_x to mean the number of splits, so we need to
                                     // cancel out the divide by wg_denoms[0].
-                                    sizeof(vk_flash_attn_push_constants), &pc, { workgroups_x * pipeline->wg_denoms[0], workgroups_y, workgroups_z });
+                                    pc, { workgroups_x * pipeline->wg_denoms[0], workgroups_y, workgroups_z });
 
         ggml_vk_sync_buffers(subctx);
         const std::array<uint32_t, 3> pc2 = { D, (uint32_t)ne1, split_k };
@@ -6121,7 +6194,7 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
                                         vk_subbuffer{ctx->prealloc_split_k, 0, VK_WHOLE_SIZE},
                                         vk_subbuffer{d_D, d_buf_offset, VK_WHOLE_SIZE},
                                     },
-                                    pc2.size() * uint32_t{sizeof(uint32_t)}, pc2.data(), { (uint32_t)ne1, 1, 1 });
+                                    pc2, { (uint32_t)ne1, 1, 1 });
     } else {
         ggml_vk_dispatch_pipeline(ctx, subctx, pipeline,
                                     {
@@ -6131,7 +6204,7 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
                                         vk_subbuffer{d_M, m_buf_offset, VK_WHOLE_SIZE},
                                         vk_subbuffer{d_D, d_buf_offset, VK_WHOLE_SIZE},
                                     },
-                                    sizeof(vk_flash_attn_push_constants), &pc, { workgroups_x, workgroups_y, workgroups_z });
+                                    pc, { workgroups_x, workgroups_y, workgroups_z });
     }
 }
 
@@ -6392,6 +6465,11 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
             return ctx->device->pipeline_timestep_embedding_f32;
         }
         return nullptr;
+    case GGML_OP_CONV_TRANSPOSE_1D:
+        if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
+            return ctx->device->pipeline_conv_transpose_1d_f32;
+        }
+        return nullptr;
     case GGML_OP_POOL_2D:
         if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
             return ctx->device->pipeline_pool2d_f32;
@@ -6726,6 +6804,10 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
             uint32_t half_ceil = (dim + 1) / 2;
             elements = { half_ceil, (uint32_t)src0->ne[0], 1 };
         } break;
+    case GGML_OP_CONV_TRANSPOSE_1D:
+        {
+            elements = {uint32_t(src0->ne[1]), 1, 1}; // parallelize in {Cout, 1, 1}
+        } break;
     case GGML_OP_POOL_2D:
         {
             const uint32_t N = dst->ne[3];
@@ -6800,7 +6882,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
         }
 
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, subbuf_y, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, subbuf_y, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else if (op == GGML_OP_ROPE || op == GGML_OP_ROPE_BACK) {
         // Empty src2 is possible in rope, but the shader needs a buffer
         vk_subbuffer subbuf_z;
@@ -6811,26 +6893,26 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
         }
 
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, subbuf_z, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, subbuf_z, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else if (op == GGML_OP_IM2COL) {
         // im2col uses only src1 and dst buffers
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else if (op == GGML_OP_COUNT_EQUAL) {
         ggml_vk_sync_buffers(subctx);
         // count_equal assumes that destination buffer is initialized with zeroes
         ggml_vk_buffer_memset_async(subctx, d_D, d_buf_offset, 0, d_sz);
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else if (use_src2) {
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_Z, z_buf_offset, z_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_Z, z_buf_offset, z_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else if (use_src1) {
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_Y, y_buf_offset, y_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     } else {
         ggml_vk_sync_buffers(subctx);
-        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, sizeof(PC), &pc, elements);
+        ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{ d_X, x_buf_offset, x_sz }, vk_subbuffer{ d_D, d_buf_offset, d_sz } }, pc, elements);
     }
 }
 
@@ -6999,7 +7081,7 @@ static void ggml_vk_op_f32_wkv(ggml_backend_vk_context * ctx, vk_context& subctx
             vk_subbuffer{ d_srcs[4], src_offsets[4], src_sizes[4] },
             vk_subbuffer{ d_srcs[5], src_offsets[5], src_sizes[5] },
             vk_subbuffer{ d_D, dst_offset, dst_size }
-        }, sizeof(vk_op_rwkv_wkv6_push_constants), &pc, elements);
+        }, pc, elements);
     } else if (version == 7) {
         ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, {
             vk_subbuffer{ d_srcs[0], src_offsets[0], src_sizes[0] },
@@ -7010,7 +7092,7 @@ static void ggml_vk_op_f32_wkv(ggml_backend_vk_context * ctx, vk_context& subctx
             vk_subbuffer{ d_srcs[5], src_offsets[5], src_sizes[5] },
             vk_subbuffer{ d_srcs[6], src_offsets[6], src_sizes[6] },
             vk_subbuffer{ d_D, dst_offset, dst_size }
-        }, sizeof(vk_op_rwkv_wkv7_push_constants), &pc, elements);
+        }, pc, elements);
     } else {
         // shouldn't happen
         GGML_ASSERT(false);
@@ -7147,7 +7229,7 @@ static void ggml_vk_op_f32_opt_step_adamw(ggml_backend_vk_context * ctx, vk_cont
         vk_subbuffer{ d_GM, gm_offset, gm_size },
         vk_subbuffer{ d_GV, gv_offset, gv_size },
         vk_subbuffer{ d_P, p_offset, p_size },
-    }, sizeof(vk_op_push_constants), &pc, elements);
+    }, pc, elements);
 }
 
 static void ggml_vk_opt_step_adamw(ggml_backend_vk_context * ctx, vk_context& subctx, ggml_tensor * dst, bool dryrun = false) {
@@ -7529,6 +7611,37 @@ static void ggml_vk_timestep_embedding(ggml_backend_vk_context * ctx, vk_context
     }, dryrun);
 }
 
+static void ggml_vk_conv_transpose_1d(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
+    // src0: (K, Cout, Cin, 1) -- kernel
+    // src1: (L, Cin, 1, 1) -- input
+    // dst: (*, Cout, 1, 1)
+
+    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+    GGML_ASSERT(src1->type == GGML_TYPE_F32);
+    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    GGML_ASSERT(nb00 == sizeof(float));
+    GGML_ASSERT(nb10 == sizeof(float));
+
+    const int32_t s0 = dst->op_params[0];
+
+    vk_op_conv_transpose_1d_push_constants p{};
+    p.Cout = static_cast<uint32_t>(ne01);
+    p.Cin = static_cast<uint32_t>(ne02);
+    p.K = static_cast<uint32_t>(ne00);
+    p.L = static_cast<uint32_t>(ne10);
+    p.KL = static_cast<uint32_t>(ne0);
+    p.nb01 = static_cast<uint32_t>(nb01 / nb00);
+    p.nb02 = static_cast<uint32_t>(nb02 / nb00);
+    p.nb11 = static_cast<uint32_t>(nb11 / nb10);
+    p.nb1 = static_cast<uint32_t>(nb1 / nb0);
+    p.s0 = static_cast<uint32_t>(s0);
+
+    ggml_vk_op_f32(ctx, subctx, src0, src1, nullptr, dst, GGML_OP_CONV_TRANSPOSE_1D, std::move(p), dryrun);
+}
+
 static void ggml_vk_pool_2d(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
     uint32_t op = static_cast<uint32_t>(dst->op_params[0]);
     const int32_t k1 = dst->op_params[1];
@@ -8005,7 +8118,7 @@ static void ggml_vk_test_dequant(ggml_backend_vk_context * ctx, size_t ne, ggml_
     vk_context subctx = ggml_vk_create_context(ctx, ctx->device->compute_queue);
     ggml_vk_ctx_begin(ctx->device, subctx);
     const std::vector<uint32_t> pc = { 1, (uint32_t)ne, (uint32_t)ne, (uint32_t)ne, (uint32_t)ne };
-    ggml_vk_dispatch_pipeline(ctx, subctx, p, { vk_subbuffer{ qx_buf, 0, qx_sz }, vk_subbuffer{ x_buf, 0, x_sz_f16 } }, pc.size() * sizeof(int), pc.data(), { (uint32_t)ne, 1, 1});
+    ggml_vk_dispatch_pipeline(ctx, subctx, p, { vk_subbuffer{ qx_buf, 0, qx_sz }, vk_subbuffer{ x_buf, 0, x_sz_f16 } }, pc, { (uint32_t)ne, 1, 1});
     ggml_vk_ctx_end(subctx);
 
     auto begin = std::chrono::high_resolution_clock::now();
@@ -8600,6 +8713,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
     case GGML_OP_COUNT_EQUAL:
     case GGML_OP_IM2COL:
     case GGML_OP_TIMESTEP_EMBEDDING:
+    case GGML_OP_CONV_TRANSPOSE_1D:
     case GGML_OP_POOL_2D:
     case GGML_OP_CONV_2D_DW:
     case GGML_OP_RWKV_WKV6:
@@ -8664,6 +8778,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
         case GGML_OP_COUNT_EQUAL:
         case GGML_OP_IM2COL:
         case GGML_OP_TIMESTEP_EMBEDDING:
+        case GGML_OP_CONV_TRANSPOSE_1D:
         case GGML_OP_POOL_2D:
         case GGML_OP_CONV_2D_DW:
         case GGML_OP_LEAKY_RELU:
@@ -8835,6 +8950,10 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
     case GGML_OP_TIMESTEP_EMBEDDING:
         ggml_vk_timestep_embedding(ctx, compute_ctx, src0, node, dryrun);
 
+        break;
+    case GGML_OP_CONV_TRANSPOSE_1D:
+        ggml_vk_conv_transpose_1d(ctx, compute_ctx, src0, src1, node, dryrun);
+
         break;
     case GGML_OP_POOL_2D:
         ggml_vk_pool_2d(ctx, compute_ctx, src0, node, dryrun);
@@ -8963,6 +9082,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_tensor *
     case GGML_OP_COUNT_EQUAL:
     case GGML_OP_IM2COL:
     case GGML_OP_TIMESTEP_EMBEDDING:
+    case GGML_OP_CONV_TRANSPOSE_1D:
     case GGML_OP_POOL_2D:
     case GGML_OP_CONV_2D_DW:
     case GGML_OP_RWKV_WKV6:
@@ -9513,8 +9633,8 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
             if (ctx->device->query_pool) {
                 ctx->device->device.destroyQueryPool(ctx->device->query_pool);
             }
-            VkQueryPoolCreateInfo query_create_info = { VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO };
-            query_create_info.queryType = VK_QUERY_TYPE_TIMESTAMP;
+            vk::QueryPoolCreateInfo query_create_info;
+            query_create_info.queryType = vk::QueryType::eTimestamp;
             query_create_info.queryCount = cgraph->n_nodes + 100;
             ctx->device->query_pool = ctx->device->device.createQueryPool(query_create_info);
             ctx->device->num_queries = query_create_info.queryCount;
@@ -9600,7 +9720,7 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
 
         // Get the results and pass them to the logger
         std::vector<uint64_t> timestamps(cgraph->n_nodes + 1);
-        ctx->device->device.getQueryPoolResults(ctx->device->query_pool, 0, cgraph->n_nodes + 1, (cgraph->n_nodes + 1)*sizeof(uint64_t), timestamps.data(), sizeof(uint64_t), vk::QueryResultFlagBits::e64 | vk::QueryResultFlagBits::eWait);
+        VK_CHECK(ctx->device->device.getQueryPoolResults(ctx->device->query_pool, 0, cgraph->n_nodes + 1, (cgraph->n_nodes + 1)*sizeof(uint64_t), timestamps.data(), sizeof(uint64_t), vk::QueryResultFlagBits::e64 | vk::QueryResultFlagBits::eWait), "get timestamp results");
         for (int i = 0; i < cgraph->n_nodes; i++) {
             if (!ggml_vk_is_empty(cgraph->nodes[i])) {
                 ctx->device->perf_logger->log_timing(cgraph->nodes[i], uint64_t((timestamps[i+1] - timestamps[i]) * ctx->device->properties.limits.timestampPeriod));
@@ -10024,6 +10144,8 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
         case GGML_OP_LEAKY_RELU:
         case GGML_OP_OPT_STEP_ADAMW:
             return true;
+        case GGML_OP_CONV_TRANSPOSE_1D:
+            return op->src[0]->type == GGML_TYPE_F32 && op->src[1]->type == GGML_TYPE_F32;
         default:
             return false;
     }
@@ -10170,8 +10292,9 @@ static bool ggml_vk_instance_portability_enumeration_ext_available(const std::ve
 static bool ggml_vk_khr_cooperative_matrix_support(const vk::PhysicalDeviceProperties& props, const vk::PhysicalDeviceDriverProperties& driver_props, vk_device_architecture arch) {
     switch (props.vendorID) {
     case VK_VENDOR_ID_INTEL:
-        // Intel drivers don't support coopmat properly yet
-        return false;
+        // Only allowing Xe2 GPU at the moment since Xe2 GPU can gain significant performance boost,
+        // while some older hardware (ex. Arc A770) has performance regressions
+        return arch == vk_device_architecture::INTEL_XE2;
     case VK_VENDOR_ID_AMD:
         if (driver_props.driverID == vk::DriverId::eAmdProprietary || driver_props.driverID == vk::DriverId::eAmdOpenSource) {
             // Workaround for AMD proprietary driver reporting support on all GPUs
@@ -10515,6 +10638,11 @@ static void ggml_vk_check_results_0(ggml_tensor * tensor) {
         const int32_t dim = tensor->op_params[0];
         const int32_t max_period = tensor->op_params[1];
         tensor_clone = ggml_timestep_embedding(ggml_ctx, src_clone[0], dim, max_period);
+    } else if (tensor->op == GGML_OP_CONV_TRANSPOSE_1D){
+        const int32_t s0 = tensor->op_params[0];
+        const int32_t p0 = tensor->op_params[1];
+        const int32_t d0 = tensor->op_params[2];
+        tensor_clone = ggml_conv_transpose_1d(ggml_ctx, src_clone[0], src_clone[1], s0, p0, d0);
     } else if (tensor->op == GGML_OP_POOL_2D) {
         enum ggml_op_pool op = static_cast<ggml_op_pool>(tensor->op_params[0]);
         const int32_t k0 = tensor->op_params[1];

ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp

@@ -0,0 +1,98 @@
+#version 450
+
+#include "types.comp"
+
+layout (binding = 0) readonly buffer A {A_TYPE data_a[];};   // src0 - kernel:    [K, Cout, Cin]
+layout (binding = 1) readonly buffer B {B_TYPE data_b[];};   // src1 - input:     [L, Cin]
+layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};     // dst - result      [KL, Cout]
+
+layout(local_size_x = 128 , local_size_y = 1, local_size_z = 1) in;
+
+layout (push_constant) uniform parameter {
+    uint32_t Cout;
+    uint32_t Cin;
+    uint32_t K;
+    uint32_t L;
+    uint32_t KL;
+
+    uint32_t nb01;
+    uint32_t nb02;
+    uint32_t nb11;
+    uint32_t nb1;
+
+    int32_t s0;
+} p;
+
+
+uint32_t Cout_idx = gl_WorkGroupID.x;
+const uint32_t bs = gl_WorkGroupSize.x;
+uint32_t tid = gl_LocalInvocationID.x;
+// Code is more straightforward if we assume it is bs*s0+K instead of (bs-1)*s0+K.
+uint32_t tmp_len = bs*p.s0+p.K;
+shared D_TYPE tmp[4096];
+
+uint splitWork(uint workSize){
+    return (bs + workSize -1) / bs;
+}
+
+void main(){
+    for(uint32_t i = 0; i < splitWork(tmp_len); i++){
+        uint32_t idx = i*bs+tid;
+        if(idx < tmp_len){
+            tmp[idx] = 0.0;
+        }
+    }
+
+    uint32_t L_blocks = splitWork(p.L);
+    for(uint32_t L_block_id = 0; L_block_id < L_blocks; L_block_id++){
+        if(L_block_id > 0){
+            barrier();
+            // Shift values in tmp to the current processing window
+            for(int i = 0; i < splitWork(tmp_len); i++){
+                uint32_t idx = i*bs+tid;
+                if(idx >= bs*p.s0 && idx < tmp_len){
+                    tmp[idx-bs*p.s0] = tmp[idx];
+                    tmp[idx] = 0.0;
+                }else if(idx >= p.K && idx < bs*p.s0){
+                    tmp[idx] = 0.0;
+                }
+            }
+        }
+        barrier();
+
+        // Save contributions of the block to tmp
+        uint32_t L_idx = L_block_id*bs + tid;
+        for(uint32_t K_idx = 0; K_idx < p.K; K_idx++){
+            D_TYPE dp = 0.0;
+            for(uint32_t Cin_idx = 0; Cin_idx < p.Cin; Cin_idx++){
+                A_TYPE elemKrn = data_a[K_idx + Cout_idx * p.nb01 + Cin_idx * p.nb02];
+                if(L_idx < p.L){
+                    B_TYPE elemInp = data_b[L_idx + Cin_idx*p.nb11];
+                    dp = fma(elemKrn, elemInp, dp);
+                }
+            }
+            tmp[tid*p.s0 + K_idx] += dp;
+            barrier();
+        }
+
+        // Save the computed values except the last block that can have different size
+        uint32_t KLb_idx = L_block_id*bs*p.s0;
+        if(L_block_id < L_blocks-1){
+            for(uint32_t s0_idx = 0; s0_idx < p.s0; s0_idx++){
+                uint32_t sh_idx = p.s0*tid+s0_idx;
+                uint32_t KL_idx = KLb_idx+sh_idx;
+                if(KL_idx < p.KL){
+                    data_d[KL_idx + Cout_idx*p.nb1] = tmp[sh_idx];
+                }
+            }
+        }
+    }
+
+    for(uint32_t i = 0; i < splitWork(tmp_len); i++){
+        uint32_t idx = i*bs+tid;
+        uint32_t KL_idx = (L_blocks-1)*bs*p.s0+idx;
+        if(KL_idx < p.KL){
+            data_d[KL_idx + Cout_idx*p.nb1] = tmp[idx];
+        }
+    }
+}

ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp

@@ -622,6 +622,8 @@ void process_shaders() {
 
     string_to_spv("timestep_embedding_f32", "timestep_embedding.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
 
+    string_to_spv("conv_transpose_1d_f32", "conv_transpose_1d.comp", {{"A_TYPE", "float"},  {"B_TYPE", "float"}, {"D_TYPE", "float"}});
+
     string_to_spv("pool2d_f32", "pool2d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
 
     string_to_spv("rwkv_wkv6_f32", "wkv6.comp", merge_maps(base_dict, {{"A_TYPE", "float"}}));

gguf-py/gguf/gguf_writer.py

@@ -935,6 +935,9 @@ class GGUFWriter:
     def add_eom_token_id(self, id: int) -> None:
         self.add_uint32(Keys.Tokenizer.EOM_ID, id)
 
+    def add_classifier_output_labels(self, labels: Sequence[str]) -> None:
+        self.add_array(Keys.Classifier.OUTPUT_LABELS.format(arch=self.arch), labels)
+
     # for vision models
 
     def add_clip_has_vision_encoder(self, value: bool) -> None:

include/llama.h

@@ -61,7 +61,10 @@ extern "C" {
     struct llama_model;
     struct llama_context;
     struct llama_sampler;
-    struct llama_kv_cache;
+
+    typedef struct llama_memory_i * llama_memory_t;
+
+    struct llama_kv_cache; // DEPRECATED (use llama_memory instead)
 
     typedef int32_t llama_pos;
     typedef int32_t llama_token;
@@ -493,9 +496,11 @@ extern "C" {
     DEPRECATED(LLAMA_API int32_t llama_n_vocab    (const struct llama_vocab * vocab), "use llama_vocab_n_tokens instead");
 
     LLAMA_API const struct llama_model * llama_get_model   (const struct llama_context * ctx);
-    LLAMA_API    struct llama_kv_cache * llama_get_kv_self (      struct llama_context * ctx);
+    LLAMA_API           llama_memory_t   llama_get_memory  (const struct llama_context * ctx);
     LLAMA_API  enum llama_pooling_type   llama_pooling_type(const struct llama_context * ctx); // TODO: rename to llama_get_pooling_type
 
+    DEPRECATED(LLAMA_API struct llama_kv_cache * llama_get_kv_self(struct llama_context * ctx), "use llama_get_memory instead");
+
     LLAMA_API const struct llama_vocab * llama_model_get_vocab(const struct llama_model * model);
     LLAMA_API enum llama_rope_type       llama_model_rope_type(const struct llama_model * model);
 
@@ -509,6 +514,13 @@ extern "C" {
     // Get the model's RoPE frequency scaling factor
     LLAMA_API float llama_model_rope_freq_scale_train(const struct llama_model * model);
 
+    // Returns the number of classifier outputs (only valid for classifier models)
+    // Undefined behavior for non-classifier models
+    LLAMA_API uint32_t llama_model_n_cls_out(const struct llama_model * model);
+
+    // Returns label of classifier output by index (<n_cls_out). Returns nullptr if no label provided
+    LLAMA_API const char * llama_model_cls_label(const struct llama_model * model, uint32_t i);
+
     LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_vocab * vocab);
 
     LLAMA_API int32_t llama_vocab_n_tokens(const struct llama_vocab * vocab);
@@ -609,7 +621,81 @@ extern "C" {
                          int32_t   il_end);
 
     //
-    // KV cache
+    // Memory
+    //
+
+    // Clear the memory contents
+    // If data == true, the data buffers will also be cleared together with the metadata
+    LLAMA_API void llama_memory_clear(
+            llama_memory_t mem,
+                      bool data);
+
+    // Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
+    // Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
+    // seq_id < 0 : match any sequence
+    // p0 < 0     : [0,  p1]
+    // p1 < 0     : [p0, inf)
+    LLAMA_API bool llama_memory_seq_rm(
+            llama_memory_t mem,
+              llama_seq_id seq_id,
+                 llama_pos p0,
+                 llama_pos p1);
+
+    // Copy all tokens that belong to the specified sequence to another sequence
+    // p0 < 0 : [0,  p1]
+    // p1 < 0 : [p0, inf)
+    LLAMA_API void llama_memory_seq_cp(
+            llama_memory_t mem,
+              llama_seq_id seq_id_src,
+              llama_seq_id seq_id_dst,
+                 llama_pos p0,
+                 llama_pos p1);
+
+    // Removes all tokens that do not belong to the specified sequence
+    LLAMA_API void llama_memory_seq_keep(
+            llama_memory_t mem,
+              llama_seq_id seq_id);
+
+    // Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
+    // p0 < 0 : [0,  p1]
+    // p1 < 0 : [p0, inf)
+    LLAMA_API void llama_memory_seq_add(
+            llama_memory_t mem,
+              llama_seq_id seq_id,
+                 llama_pos p0,
+                 llama_pos p1,
+                 llama_pos delta);
+
+    // Integer division of the positions by factor of `d > 1`
+    // p0 < 0 : [0,  p1]
+    // p1 < 0 : [p0, inf)
+    LLAMA_API void llama_memory_seq_div(
+            llama_memory_t mem,
+              llama_seq_id seq_id,
+                 llama_pos p0,
+                 llama_pos p1,
+                       int d);
+
+    // Returns the smallest position present in the memory for the specified sequence
+    // This is typically non-zero only for SWA caches
+    // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the memory
+    // Return -1 if the sequence is empty
+    LLAMA_API llama_pos llama_memory_seq_pos_min(
+            llama_memory_t mem,
+              llama_seq_id seq_id);
+
+    // Returns the largest position present in the memory for the specified sequence
+    // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the memory
+    // Return -1 if the sequence is empty
+    LLAMA_API llama_pos llama_memory_seq_pos_max(
+            llama_memory_t mem,
+              llama_seq_id seq_id);
+
+    // Check if the memory supports shifting
+    LLAMA_API bool llama_memory_can_shift(llama_memory_t mem);
+
+    //
+    // KV cache for self-attention (TODO: deprecate in favor of llama_memory)
     //
 
     // Returns the number of tokens in the KV cache (slow, use only for debug)
@@ -622,86 +708,95 @@ extern "C" {
                "Use llama_kv_self_seq_pos_max() and llama_kv_self_seq_pos_min() instead (https://github.com/ggml-org/llama.cpp/issues/13793)");
 
     // Clear the KV cache - both cell info is erased and KV data is zeroed
-    LLAMA_API void llama_kv_self_clear(
-            struct llama_context * ctx);
+    DEPRECATED(LLAMA_API void llama_kv_self_clear(
+                struct llama_context * ctx),
+            "Use llama_memory_clear() instead");
 
     // Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
     // Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
     // seq_id < 0 : match any sequence
     // p0 < 0     : [0,  p1]
     // p1 < 0     : [p0, inf)
-    LLAMA_API bool llama_kv_self_seq_rm(
+    DEPRECATED(LLAMA_API bool llama_kv_self_seq_rm(
             struct llama_context * ctx,
                     llama_seq_id   seq_id,
                        llama_pos   p0,
-                       llama_pos   p1);
+                       llama_pos   p1),
+            "Use llama_memory_seq_rm() instead");
 
     // Copy all tokens that belong to the specified sequence to another sequence
     // Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence
     // p0 < 0 : [0,  p1]
     // p1 < 0 : [p0, inf)
-    LLAMA_API void llama_kv_self_seq_cp(
+    DEPRECATED(LLAMA_API void llama_kv_self_seq_cp(
             struct llama_context * ctx,
                     llama_seq_id   seq_id_src,
                     llama_seq_id   seq_id_dst,
                        llama_pos   p0,
-                       llama_pos   p1);
+                       llama_pos   p1),
+            "Use llama_memory_seq_cp() instead");
 
     // Removes all tokens that do not belong to the specified sequence
-    LLAMA_API void llama_kv_self_seq_keep(
+    DEPRECATED(LLAMA_API void llama_kv_self_seq_keep(
             struct llama_context * ctx,
-                    llama_seq_id   seq_id);
+                    llama_seq_id   seq_id),
+            "Use llama_memory_seq_keep() instead");
 
     // Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
     // If the KV cache is RoPEd, the KV data is updated accordingly:
     //   - lazily on next llama_decode()
     // p0 < 0 : [0,  p1]
     // p1 < 0 : [p0, inf)
-    LLAMA_API void llama_kv_self_seq_add(
+    DEPRECATED(LLAMA_API void llama_kv_self_seq_add(
             struct llama_context * ctx,
                     llama_seq_id   seq_id,
                        llama_pos   p0,
                        llama_pos   p1,
-                       llama_pos   delta);
+                       llama_pos   delta),
+            "Use llama_memory_seq_add() instead");
 
     // Integer division of the positions by factor of `d > 1`
     // If the KV cache is RoPEd, the KV data is updated accordingly:
     //   - lazily on next llama_decode()
     // p0 < 0 : [0,  p1]
     // p1 < 0 : [p0, inf)
-    LLAMA_API void llama_kv_self_seq_div(
+    DEPRECATED(void llama_kv_self_seq_div(
             struct llama_context * ctx,
                     llama_seq_id   seq_id,
                        llama_pos   p0,
                        llama_pos   p1,
-                             int   d);
+                             int   d),
+            "Use llama_memory_seq_div() instead");
 
     // Returns the smallest position present in the KV cache for the specified sequence
     // This is typically non-zero only for SWA caches
     // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the KV cache
     // Return -1 if the sequence is empty
-    LLAMA_API llama_pos llama_kv_self_seq_pos_min(
+    DEPRECATED(LLAMA_API llama_pos llama_kv_self_seq_pos_min(
             struct llama_context * ctx,
-                    llama_seq_id   seq_id);
+                    llama_seq_id   seq_id),
+            "Use llama_memory_seq_pos_min() instead");
 
     // Returns the largest position present in the KV cache for the specified sequence
     // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the KV cache
     // Return -1 if the sequence is empty
-    LLAMA_API llama_pos llama_kv_self_seq_pos_max(
+    DEPRECATED(LLAMA_API llama_pos llama_kv_self_seq_pos_max(
             struct llama_context * ctx,
-                    llama_seq_id   seq_id);
+                    llama_seq_id   seq_id),
+            "Use llama_memory_seq_pos_max() instead");
 
     // Defragment the KV cache
     // This will be applied:
     //   - lazily on next llama_decode()
-    LLAMA_API DEPRECATED(void llama_kv_self_defrag(struct llama_context * ctx),
+    DEPRECATED(LLAMA_API void llama_kv_self_defrag(struct llama_context * ctx),
             "simply remove this call, the context will automatically decide when to do a defragmentation based on 'defrag_thold'");
 
     // Check if the context supports KV cache shifting
-    LLAMA_API bool llama_kv_self_can_shift(const struct llama_context * ctx);
+    DEPRECATED(LLAMA_API bool llama_kv_self_can_shift(const struct llama_context * ctx),
+            "use llama_memory_can_shift() instead");
 
     // Apply the KV cache updates (such as K-shifts, defragmentation, etc.)
-    LLAMA_API DEPRECATED(void llama_kv_self_update(struct llama_context * ctx),
+    DEPRECATED(LLAMA_API void llama_kv_self_update(struct llama_context * ctx),
             "simply remove this call, updates are applied lazily on the next llama_decode()");
 
     //
@@ -709,7 +804,7 @@ extern "C" {
     //
 
     // Returns the *actual* size in bytes of the state
-    // (logits, embedding and kv_cache)
+    // (logits, embedding and memory)
     // Only use when saving the state, not when restoring it, otherwise the size may be too small.
     LLAMA_API size_t llama_state_get_size(struct llama_context * ctx);
     LLAMA_API DEPRECATED(size_t llama_get_state_size(struct llama_context * ctx),
@@ -765,12 +860,12 @@ extern "C" {
                           size_t   n_token_count),
         "use llama_state_save_file instead");
 
-    // Get the exact size needed to copy the KV cache of a single sequence
+    // Get the exact size needed to copy the state of a single sequence
     LLAMA_API size_t llama_state_seq_get_size(
             struct llama_context * ctx,
                     llama_seq_id   seq_id);
 
-    // Copy the KV cache of a single sequence into the specified buffer
+    // Copy the state of a single sequence into the specified buffer
     LLAMA_API size_t llama_state_seq_get_data(
             struct llama_context * ctx,
                          uint8_t * dst,
@@ -836,16 +931,16 @@ extern "C" {
     // For encode-decoder contexts, processes the batch using the encoder.
     // Can store the encoder output internally for later use by the decoder's cross-attention layers.
     //   0 - success
-    // < 0 - error. the KV cache state is restored to the state before this call
+    // < 0 - error. the memory state is restored to the state before this call
     LLAMA_API int32_t llama_encode(
             struct llama_context * ctx,
               struct llama_batch   batch);
 
     // Process a batch of tokens.
-    // Requires KV cache.
+    // Requires the context to have a memory.
     // For encode-decoder contexts, processes the batch using the decoder.
     // Positive return values does not mean a fatal error, but rather a warning.
-    // Upon non-zero return values, the KV cache state is restored to the state before this call
+    // Upon non-zero return values, the memory state is restored to the state before this call
     //    0 - success
     //    1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
     //    2 - aborted
@@ -916,7 +1011,7 @@ extern "C" {
 
     // Get the embeddings for a sequence id
     // Returns NULL if pooling_type is LLAMA_POOLING_TYPE_NONE
-    // when pooling_type == LLAMA_POOLING_TYPE_RANK, returns float[1] with the rank of the sequence
+    // when pooling_type == LLAMA_POOLING_TYPE_RANK, returns float[n_cls_out] with the rank(s) of the sequence
     // otherwise: float[n_embd] (1-dimensional)
     LLAMA_API float * llama_get_embeddings_seq(struct llama_context * ctx, llama_seq_id seq_id);
 

src/CMakeLists.txt

@@ -20,7 +20,6 @@ add_library(llama
             llama-hparams.cpp
             llama-impl.cpp
             llama-io.cpp
-            llama-kv-cache.cpp
             llama-kv-cache-unified.cpp
             llama-kv-cache-unified-iswa.cpp
             llama-kv-cache-recurrent.cpp

src/llama-arch.cpp

@@ -200,7 +200,6 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_TOKENIZER_HF_JSON,              "tokenizer.huggingface.json"              },
     { LLM_KV_TOKENIZER_RWKV,                 "tokenizer.rwkv.world"                    },
     { LLM_KV_TOKENIZER_CHAT_TEMPLATE,        "tokenizer.chat_template"                 },
-    { LLM_KV_TOKENIZER_CHAT_TEMPLATE_N,      "tokenizer.chat_template.%s"              },
     { LLM_KV_TOKENIZER_FIM_PRE_ID,           "tokenizer.ggml.fim_pre_token_id"         },
     { LLM_KV_TOKENIZER_FIM_SUF_ID,           "tokenizer.ggml.fim_suf_token_id"         },
     { LLM_KV_TOKENIZER_FIM_MID_ID,           "tokenizer.ggml.fim_mid_token_id"         },
@@ -1707,8 +1706,14 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
 LLM_KV::LLM_KV(llm_arch arch, const char * suffix) : arch(arch), suffix(suffix) {}
 
 std::string LLM_KV::operator()(llm_kv kv) const {
-    return suffix ? ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch), suffix)
-        : ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
+    std::string name = ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
+
+    if (suffix != nullptr) {
+        name += ".";
+        name += suffix;
+    }
+
+    return name;
 }
 
 std::string LLM_TN_IMPL::str() const {

src/llama-arch.h

@@ -196,7 +196,6 @@ enum llm_kv {
     LLM_KV_TOKENIZER_HF_JSON,
     LLM_KV_TOKENIZER_RWKV,
     LLM_KV_TOKENIZER_CHAT_TEMPLATE,
-    LLM_KV_TOKENIZER_CHAT_TEMPLATE_N,
     LLM_KV_TOKENIZER_FIM_PRE_ID,
     LLM_KV_TOKENIZER_FIM_SUF_ID,
     LLM_KV_TOKENIZER_FIM_MID_ID,

src/llama-context.cpp

@@ -2,9 +2,9 @@
 
 #include "llama-impl.h"
 #include "llama-io.h"
+#include "llama-memory.h"
 #include "llama-mmap.h"
 #include "llama-model.h"
-#include "llama-kv-cache.h"
 
 #include <cinttypes>
 #include <cstring>
@@ -123,7 +123,7 @@ llama_context::llama_context(
                 __func__, n_ctx_per_seq, hparams.n_ctx_train);
     }
 
-    if (!params.swa_full && cparams.n_seq_max > 1) {
+    if (!params.swa_full && cparams.n_seq_max > 1 && hparams.is_swa_any()) {
         LLAMA_LOG_WARN("%s: requested n_seq_max (%u) > 1, but swa_full is not enabled -- performance may be degraded: %s\n",
                 __func__, cparams.n_seq_max, "https://github.com/ggml-org/llama.cpp/pull/13845#issuecomment-2924800573");
     }
@@ -277,10 +277,9 @@ llama_context::llama_context(
         int n_nodes_tg  = -1;
 
         // simulate full KV cache
-        llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
 
-        const auto kv_state = kv_self->init_full();
-        if (!kv_state) {
+        const auto mstate = memory->init_full();
+        if (!mstate) {
             throw std::runtime_error("failed to initialize KV cache");
         }
 
@@ -288,7 +287,7 @@ llama_context::llama_context(
 
         // reserve pp graph first so that buffers are only allocated once
         {
-            auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, kv_state.get());
+            auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, mstate.get());
             if (!gf) {
                 throw std::runtime_error("failed to allocate compute pp buffers");
             }
@@ -299,7 +298,7 @@ llama_context::llama_context(
 
         // reserve with tg graph to get the number of splits and nodes
         {
-            auto * gf = graph_reserve(1, 1, 1, kv_state.get());
+            auto * gf = graph_reserve(1, 1, 1, mstate.get());
             if (!gf) {
                 throw std::runtime_error("failed to allocate compute tg buffers");
             }
@@ -310,7 +309,7 @@ llama_context::llama_context(
 
         // reserve again with pp graph to avoid ggml-alloc reallocations during inference
         {
-            auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, kv_state.get());
+            auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, mstate.get());
             if (!gf) {
                 throw std::runtime_error("failed to allocate compute pp buffers");
             }
@@ -419,40 +418,68 @@ uint32_t llama_context::n_threads_batch() const {
     return cparams.n_threads_batch;
 }
 
-llama_kv_cache * llama_context::get_kv_self() {
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-    return kv_self;
+llama_memory_t llama_context::get_memory() const {
+    return memory.get();
 }
 
-const llama_kv_cache * llama_context::get_kv_self() const {
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-    return kv_self;
+// deprecated
+void llama_context::kv_self_defrag_sched() {
+    if (!memory) {
+        return;
+    }
+
+    memory_force_optimize = true;
 }
 
-bool llama_context::kv_self_update() {
+// deprecated
+bool llama_context::kv_self_update(bool optimize) {
     if (!memory) {
         return false;
     }
 
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
+    {
+        // TODO: remove in the future
+        optimize |= memory_force_optimize;
+        memory_force_optimize = false;
 
-    if (!kv_self->update(*this)) {
-        // no updates have been performed
-        return false;
+        const auto mstate = memory->init_update(this, optimize);
+        switch (mstate->get_status()) {
+            case LLAMA_MEMORY_STATUS_SUCCESS:
+                {
+                    // noop
+                } break;
+            case LLAMA_MEMORY_STATUS_NO_UPDATE:
+                {
+                    // no updates need to be performed
+                    return false;
+                }
+            case LLAMA_MEMORY_STATUS_FAILED_PREPARE:
+            case LLAMA_MEMORY_STATUS_FAILED_COMPUTE:
+                {
+                    LLAMA_LOG_ERROR("%s: failed to prepare memory update\n", __func__);
+                    return false;
+                }
+        }
+
+        if (!mstate->apply()) {
+            LLAMA_LOG_ERROR("%s: failed to apply memory update\n", __func__);
+        }
     }
 
-    // if the KV cache did any computation, we have to reserve a new worst-case graph
-    const auto kv_state = kv_self->init_full();
-    if (!kv_state) {
-        throw std::runtime_error("failed to initialize KV cache");
-    }
+    // if the memory module did any computation, we have to reserve a new worst-case graph
+    {
+        const auto mstate = memory->init_full();
+        if (!mstate) {
+            throw std::runtime_error("failed to initialize memory state");
+        }
 
-    const uint32_t n_seqs   = cparams.n_seq_max;
-    const uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
+        const uint32_t n_seqs   = cparams.n_seq_max;
+        const uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
 
-    auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, kv_state.get());
-    if (!gf) {
-        LLAMA_LOG_ERROR("%s: failed to reserve graph after the KV cache update\n", __func__);
+        auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, mstate.get());
+        if (!gf) {
+            LLAMA_LOG_ERROR("%s: failed to reserve graph after the memory update\n", __func__);
+        }
     }
 
     return true;
@@ -814,16 +841,17 @@ int llama_context::encode(llama_batch & inp_batch) {
                 } break;
             case LLAMA_POOLING_TYPE_RANK:
                 {
-                    // extract the rerank score - a single float per sequence
+                    // extract the rerank score - n_cls_out floats per sequence
                     auto & embd_seq_out = embd_seq;
+                    const uint32_t n_cls_out = hparams.n_cls_out;
 
                     for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
                         const llama_seq_id seq_id = ubatch.seq_id[s][0];
                         if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
                             continue;
                         }
-                        embd_seq_out[seq_id].resize(1);
-                        ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (seq_id)*sizeof(float), sizeof(float));
+                        embd_seq_out[seq_id].resize(n_cls_out);
+                        ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_cls_out*seq_id)*sizeof(float), n_cls_out*sizeof(float));
                     }
                 } break;
             case LLAMA_POOLING_TYPE_UNSPECIFIED:
@@ -880,10 +908,8 @@ int llama_context::decode(llama_batch & inp_batch) {
         }
     }
 
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
     // temporary allocate memory for the input batch if needed
-    llama_batch_allocr batch_allocr(inp_batch, inp_batch.pos ? -1 : kv_self->seq_pos_max(0) + 1);
+    llama_batch_allocr batch_allocr(inp_batch, inp_batch.pos ? -1 : memory->seq_pos_max(0) + 1);
 
     const llama_batch & batch = batch_allocr.batch;
 
@@ -940,42 +966,49 @@ int llama_context::decode(llama_batch & inp_batch) {
         n_outputs_all = 1;
     }
 
+    bool did_optimize = false;
+
     // handle any pending defrags/shifts
-    kv_self_update();
+    kv_self_update(false);
 
-    llama_memory_state_ptr kv_state;
-
-    bool did_defrag = false;
+    llama_memory_state_ptr mstate;
 
     while (true) {
-        kv_state = kv_self->init_batch(batch, cparams.n_ubatch, embd_pooled, /* logits_all */ n_outputs_all == n_tokens_all);
-        if (!kv_state) {
+        mstate = memory->init_batch(batch, cparams.n_ubatch, embd_pooled, /* logits_all */ n_outputs_all == n_tokens_all);
+        if (!mstate) {
             return -2;
         }
 
-        switch (kv_state->get_status()) {
+        switch (mstate->get_status()) {
             case LLAMA_MEMORY_STATUS_SUCCESS:
                 {
                 } break;
+            case LLAMA_MEMORY_STATUS_NO_UPDATE:
+                {
+                    LLAMA_LOG_ERROR("%s: unexpected memory state status: %d\n", __func__, mstate->get_status());
+
+                    return -2;
+                }
             case LLAMA_MEMORY_STATUS_FAILED_PREPARE:
                 {
-                    if (!did_defrag) {
-                        did_defrag = true;
+                    if (!did_optimize) {
+                        did_optimize = true;
 
-                        kv_self->defrag_sched(-1.0f);
-                        if (kv_self_update()) {
-                            LLAMA_LOG_DEBUG("%s: failed to init batch of size %d, retrying after defrag\n", __func__, batch.n_tokens);
+                        if (kv_self_update(true)) {
+                            LLAMA_LOG_DEBUG("%s: retrying batch size %d after cache optimization\n", __func__, batch.n_tokens);
 
                             continue;
                         }
                     }
 
-                    LLAMA_LOG_WARN("%s: failed to find KV cache slot for batch of size %d\n", __func__, batch.n_tokens);
+                    LLAMA_LOG_WARN("%s: failed to find a memory slot for batch of size %d\n", __func__, batch.n_tokens);
 
                     return 1;
                 }
             case LLAMA_MEMORY_STATUS_FAILED_COMPUTE:
                 {
+                    LLAMA_LOG_ERROR("%s: compute failed while preparing batch of size %d\n", __func__, batch.n_tokens);
+
                     return -2;
                 }
         }
@@ -992,7 +1025,7 @@ int llama_context::decode(llama_batch & inp_batch) {
     int64_t n_outputs_prev = 0;
 
     do {
-        const auto & ubatch = kv_state->get_ubatch();
+        const auto & ubatch = mstate->get_ubatch();
 
         // count the outputs in this u_batch
         {
@@ -1015,11 +1048,14 @@ int llama_context::decode(llama_batch & inp_batch) {
         ggml_backend_sched_set_eval_callback(sched.get(), cparams.cb_eval, cparams.cb_eval_user_data);
 
         ggml_status status;
-        const auto res = process_ubatch(ubatch, LLM_GRAPH_TYPE_DECODER, kv_state.get(), status);
+        const auto res = process_ubatch(ubatch, LLM_GRAPH_TYPE_DECODER, mstate.get(), status);
 
         if (!res) {
             // the last ubatch failed or was aborted -> remove all positions of that ubatch from the KV cache
-            llama_pos pos_min[LLAMA_MAX_PARALLEL_SEQUENCES] = { std::numeric_limits<llama_pos>::max() };
+            llama_pos pos_min[LLAMA_MAX_PARALLEL_SEQUENCES];
+            for (int s = 0; s < LLAMA_MAX_PARALLEL_SEQUENCES; ++s) {
+                pos_min[s] = std::numeric_limits<llama_pos>::max();
+            }
 
             for (uint32_t i = 0; i < ubatch.n_tokens; ++i) {
                 const auto & seq_id = ubatch.seq_id[i][0];
@@ -1034,7 +1070,7 @@ int llama_context::decode(llama_batch & inp_batch) {
 
                 LLAMA_LOG_WARN("%s: removing KV cache entries for seq_id = %d, pos = [%d, +inf)\n", __func__, s, pos_min[s]);
 
-                llama_kv_self_seq_rm(this, s, pos_min[s], -1);
+                memory->seq_rm(s, pos_min[s], -1);
             }
 
             switch (status) {
@@ -1128,7 +1164,7 @@ int llama_context::decode(llama_batch & inp_batch) {
         }
 
         n_outputs_prev += n_outputs;
-    } while (kv_state->next());
+    } while (mstate->next());
 
     // set to total number of outputs in the batch, for use in llama_get_logits_ith
     n_outputs = n_outputs_all;
@@ -1137,7 +1173,7 @@ int llama_context::decode(llama_batch & inp_batch) {
     {
         bool sorted_output = true;
 
-        auto & out_ids = kv_state->out_ids();
+        auto & out_ids = mstate->out_ids();
 
         GGML_ASSERT(out_ids.size() == (size_t) n_outputs_all);
 
@@ -1189,11 +1225,6 @@ int llama_context::decode(llama_batch & inp_batch) {
     // wait for the computation to finish (automatically done when obtaining the model output)
     //synchronize();
 
-    // decide if we need to defrag the kv cache
-    if (cparams.defrag_thold > 0.0f) {
-        kv_self->defrag_sched(cparams.defrag_thold);
-    }
-
     // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
     // overlap with device computation.
     ggml_backend_sched_reset(sched.get());
@@ -1810,11 +1841,9 @@ size_t llama_context::state_write_data(llama_io_write_i & io) {
         }
     }
 
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
-    if (kv_self != nullptr) {
+    if (memory != nullptr) {
         LLAMA_LOG_DEBUG("%s: - writing KV self\n", __func__);
-        kv_self->state_write(io);
+        memory->state_write(io);
     }
 
     return io.n_bytes();
@@ -1901,9 +1930,7 @@ size_t llama_context::state_read_data(llama_io_read_i & io) {
     if (memory) {
         LLAMA_LOG_DEBUG("%s: - reading KV self\n", __func__);
 
-        llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
-        kv_self->state_read(io);
+        memory->state_read(io);
     }
 
     return io.n_bytes();
@@ -1913,9 +1940,7 @@ size_t llama_context::state_seq_write_data(llama_io_write_i & io, llama_seq_id s
     GGML_UNUSED(seq_id);
 
     if (memory) {
-        llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
-        kv_self->state_write(io, seq_id);
+        memory->state_write(io, seq_id);
     }
 
     return io.n_bytes();
@@ -1925,9 +1950,7 @@ size_t llama_context::state_seq_read_data(llama_io_read_i & io, llama_seq_id seq
     GGML_UNUSED(seq_id);
 
     if (memory) {
-        llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
-        kv_self->state_read(io, seq_id);
+        memory->state_read(io, seq_id);
     }
 
     return io.n_bytes();
@@ -2032,9 +2055,7 @@ void llama_context::opt_epoch_iter(
     const uint32_t n_batch  = std::min(this->n_batch(),  n_ctx);
     const uint32_t n_ubatch = std::min(this->n_ubatch(), n_batch);
 
-    llama_kv_cache * kv_self = static_cast<llama_kv_cache *>(memory.get());
-
-    kv_self->clear();
+    memory->clear(true);
 
     for (uint32_t pos_ctx = 0; pos_ctx < n_ctx; pos_ctx += n_batch) {
         batch.n_tokens = n_batch;
@@ -2057,8 +2078,8 @@ void llama_context::opt_epoch_iter(
 
         int64_t n_outputs_all = n_tokens_all;
 
-        auto kv_state = kv_self->init_batch(batch, cparams.n_ubatch, embd_pooled, /* logits_all */ true);
-        if (!kv_state || kv_state->get_status() != LLAMA_MEMORY_STATUS_SUCCESS) {
+        auto mstate = memory->init_batch(batch, cparams.n_ubatch, embd_pooled, /* logits_all */ true);
+        if (!mstate || mstate->get_status() != LLAMA_MEMORY_STATUS_SUCCESS) {
             LLAMA_LOG_ERROR("%s: could not initialize batch\n", __func__);
             break;
         }
@@ -2071,17 +2092,17 @@ void llama_context::opt_epoch_iter(
 
         uint32_t pos_batch = 0;
         do {
-            const auto & ubatch = kv_state->get_ubatch();
+            const auto & ubatch = mstate->get_ubatch();
 
             n_outputs = ubatch.n_tokens;
 
-            if (!kv_state->apply()) {
+            if (!mstate->apply()) {
                 LLAMA_LOG_ERROR("%s: failed to update the memory state\n", __func__);
                 break;
             }
 
             auto * gf = graph_init();
-            auto res = graph_build(ctx_compute.get(), gf, ubatch, LLM_GRAPH_TYPE_DEFAULT, kv_state.get());
+            auto res = graph_build(ctx_compute.get(), gf, ubatch, LLM_GRAPH_TYPE_DEFAULT, mstate.get());
 
             struct ggml_context * ctx_compute_opt;
             {
@@ -2116,7 +2137,7 @@ void llama_context::opt_epoch_iter(
             ggml_free(ctx_compute_opt);
 
             pos_batch += ubatch.n_tokens;
-        } while (kv_state->next());
+        } while (mstate->next());
     }
 }
 
@@ -2277,13 +2298,14 @@ const llama_model * llama_get_model(const llama_context * ctx) {
     return &ctx->get_model();
 }
 
+// deprecated
 llama_kv_cache * llama_get_kv_self(llama_context * ctx) {
-    return ctx->get_kv_self();
+    return dynamic_cast<llama_kv_cache *>(ctx->get_memory());
 }
 
 // deprecated
 void llama_kv_self_update(llama_context * ctx) {
-    ctx->kv_self_update();
+    ctx->kv_self_update(false);
 }
 
 enum llama_pooling_type llama_pooling_type(const llama_context * ctx) {
@@ -2398,13 +2420,118 @@ int32_t llama_apply_adapter_cvec(
     return res ? 0 : -1;
 }
 
+//
+// memory
+//
+
+llama_memory_t llama_get_memory(const struct llama_context * ctx) {
+    return ctx->get_memory();
+}
+
+void llama_memory_clear(llama_memory_t mem, bool data) {
+    if (!mem) {
+        return;
+    }
+
+    mem->clear(data);
+}
+
+bool llama_memory_seq_rm(
+        llama_memory_t mem,
+          llama_seq_id seq_id,
+             llama_pos p0,
+             llama_pos p1) {
+    if (!mem) {
+        return true;
+    }
+
+    return mem->seq_rm(seq_id, p0, p1);
+}
+
+void llama_memory_seq_cp(
+        llama_memory_t mem,
+          llama_seq_id seq_id_src,
+          llama_seq_id seq_id_dst,
+             llama_pos p0,
+             llama_pos p1) {
+    if (!mem) {
+        return;
+    }
+
+    mem->seq_cp(seq_id_src, seq_id_dst, p0, p1);
+}
+
+void llama_memory_seq_keep(
+        llama_memory_t mem,
+          llama_seq_id seq_id) {
+    if (!mem) {
+        return;
+    }
+
+    mem->seq_keep(seq_id);
+}
+
+void llama_memory_seq_add(
+        llama_memory_t mem,
+          llama_seq_id seq_id,
+             llama_pos p0,
+             llama_pos p1,
+             llama_pos delta) {
+    if (!mem) {
+        return;
+    }
+
+    mem->seq_add(seq_id, p0, p1, delta);
+}
+
+void llama_memory_seq_div(
+        llama_memory_t mem,
+          llama_seq_id seq_id,
+             llama_pos p0,
+             llama_pos p1,
+                   int d) {
+    if (!mem) {
+        return;
+    }
+
+    mem->seq_div(seq_id, p0, p1, d);
+}
+
+llama_pos llama_memory_seq_pos_min(
+        llama_memory_t mem,
+          llama_seq_id seq_id) {
+    if (!mem) {
+        return -1;
+    }
+
+    return mem->seq_pos_min(seq_id);
+}
+
+llama_pos llama_memory_seq_pos_max(
+        llama_memory_t mem,
+          llama_seq_id seq_id) {
+    if (!mem) {
+        return -1;
+    }
+
+    return mem->seq_pos_max(seq_id);
+}
+
+bool llama_memory_can_shift(llama_memory_t mem) {
+    if (!mem) {
+        return false;
+    }
+
+    return mem->get_can_shift();
+}
+
 //
 // kv cache
 //
 
 // deprecated
 int32_t llama_kv_self_n_tokens(const llama_context * ctx) {
-    const auto * kv = ctx->get_kv_self();
+    const auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return 0;
     }
@@ -2426,7 +2553,7 @@ int32_t llama_kv_self_n_tokens(const llama_context * ctx) {
 // deprecated
 // note: this is the same as above - will be removed anyway, so it's ok
 int32_t llama_kv_self_used_cells(const llama_context * ctx) {
-    const auto * kv = ctx->get_kv_self();
+    const auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return 0;
     }
@@ -2445,115 +2572,119 @@ int32_t llama_kv_self_used_cells(const llama_context * ctx) {
     return res;
 }
 
+// deprecated
 void llama_kv_self_clear(llama_context * ctx) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return;
     }
 
-    kv->clear();
+    llama_memory_clear(kv, true);
 }
 
+// deprecated
 bool llama_kv_self_seq_rm(
         llama_context * ctx,
          llama_seq_id   seq_id,
             llama_pos   p0,
             llama_pos   p1) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return true;
     }
 
-    return kv->seq_rm(seq_id, p0, p1);
+    return llama_memory_seq_rm(kv, seq_id, p0, p1);
 }
 
+// deprecated
 void llama_kv_self_seq_cp(
         llama_context * ctx,
          llama_seq_id   seq_id_src,
          llama_seq_id   seq_id_dst,
             llama_pos   p0,
             llama_pos   p1) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return;
     }
 
-    kv->seq_cp(seq_id_src, seq_id_dst, p0, p1);
+    llama_memory_seq_cp(kv, seq_id_src, seq_id_dst, p0, p1);
 }
 
+// deprecated
 void llama_kv_self_seq_keep(llama_context * ctx, llama_seq_id seq_id) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return;
     }
 
-    kv->seq_keep(seq_id);
+    llama_memory_seq_keep(kv, seq_id);
 }
 
+// deprecated
 void llama_kv_self_seq_add(
         llama_context * ctx,
          llama_seq_id   seq_id,
             llama_pos   p0,
             llama_pos   p1,
             llama_pos   delta) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return;
     }
 
-    kv->seq_add(seq_id, p0, p1, delta);
+    llama_memory_seq_add(kv, seq_id, p0, p1, delta);
 }
 
+// deprecated
 void llama_kv_self_seq_div(
         llama_context * ctx,
          llama_seq_id   seq_id,
             llama_pos   p0,
             llama_pos   p1,
                   int   d) {
-    auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return;
     }
 
-    kv->seq_div(seq_id, p0, p1, d);
+    llama_memory_seq_div(kv, seq_id, p0, p1, d);
 }
 
+// deprecated
 llama_pos llama_kv_self_seq_pos_min(llama_context * ctx, llama_seq_id seq_id) {
-    const auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return -1;
     }
 
-    return kv->seq_pos_min(seq_id);
+    return llama_memory_seq_pos_min(kv, seq_id);
 }
 
+// deprecated
 llama_pos llama_kv_self_seq_pos_max(llama_context * ctx, llama_seq_id seq_id) {
-    const auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return -1;
     }
 
-    return kv->seq_pos_max(seq_id);
+    return llama_memory_seq_pos_max(kv, seq_id);
 }
 
 // deprecated
 void llama_kv_self_defrag(llama_context * ctx) {
-    auto * kv = ctx->get_kv_self();
-    if (!kv) {
-        return;
-    }
-
     // force defrag
-    kv->defrag_sched(-1.0f);
+    ctx->kv_self_defrag_sched();
 }
 
+// deprecated
 bool llama_kv_self_can_shift(const llama_context * ctx) {
-    const auto * kv = ctx->get_kv_self();
+    auto * kv = llama_get_memory(ctx);
     if (!kv) {
         return false;
     }
 
-    return kv->get_can_shift();
+    return llama_memory_can_shift(kv);
 }
 
 // llama state API

src/llama-context.h

@@ -13,13 +13,12 @@
 #include <vector>
 
 struct llama_model;
-struct llama_kv_cache;
 
 class llama_io_read_i;
 class llama_io_write_i;
 
-class llama_memory_i;
-class llama_memory_state_i;
+struct llama_memory_i;
+struct llama_memory_state_i;
 
 struct llama_context {
     // init scheduler and compute buffers, reserve worst-case graphs
@@ -47,12 +46,12 @@ struct llama_context {
     uint32_t n_threads()       const;
     uint32_t n_threads_batch() const;
 
-          llama_kv_cache * get_kv_self();
-    const llama_kv_cache * get_kv_self() const;
+    llama_memory_t get_memory() const;
 
     // return true of the KV cache was updated
     // TODO: remove
-    bool kv_self_update();
+    bool kv_self_update(bool optimize);
+    void kv_self_defrag_sched();
 
     enum llama_pooling_type pooling_type() const;
 
@@ -231,6 +230,9 @@ private:
 
     std::unique_ptr<llama_memory_i> memory;
 
+    // TODO: temporary, until the llama_kv_self_defrag() API is removed
+    bool memory_force_optimize = false;
+
     // decode output (2-dimensional array: [n_outputs][n_vocab])
     size_t  logits_size = 0; // capacity (of floats) for logits
     float * logits      = nullptr;

src/llama-graph.cpp

@@ -250,22 +250,6 @@ void llm_graph_input_s_copy::set_input(const llama_ubatch * ubatch) {
     }
 }
 
-void llm_graph_input_s_mask::set_input(const llama_ubatch * ubatch) {
-    GGML_UNUSED(ubatch);
-
-    const int64_t n_kv = kv_state->get_n_kv();
-
-    if (s_mask) {
-        GGML_ASSERT(ggml_backend_buffer_is_host(s_mask->buffer));
-        float * data = (float *) s_mask->data;
-
-        // clear unused states
-        for (int i = 0; i < n_kv; ++i) {
-            data[i] = kv_state->s_mask(i);
-        }
-    }
-}
-
 void llm_graph_input_cross_embd::set_input(const llama_ubatch * ubatch) {
     GGML_UNUSED(ubatch);
 
@@ -659,6 +643,20 @@ ggml_tensor * llm_graph_context::build_ffn(
                 cur = ggml_mul(ctx0, x0, x1);
                 cb(cur, "ffn_mul", il);
             } break;
+        case LLM_FFN_GEGLU:
+            {
+                // Split into two equal parts
+                int64_t split_point = cur->ne[0] / 2;
+                // TODO: these conts should not be needed
+                ggml_tensor * x0 = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, split_point, cur->ne[1], cur->nb[1], 0));
+                ggml_tensor * x1 = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, split_point, cur->ne[1], cur->nb[1], split_point * ggml_element_size(cur)));
+
+                x0 = ggml_gelu(ctx0, x0);
+                cb(x0, "ffn_gelu", il);
+
+                cur = ggml_mul(ctx0, x0, x1);
+                cb(cur, "ffn_geglu", il);
+            } break;
     }
 
     if (gate && type_gate == LLM_FFN_PAR) {
@@ -769,9 +767,8 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
     cur = ggml_reshape_3d(ctx0, cur, n_embd, 1, n_tokens);
 
     if (weight_before_ffn) {
-        // TODO: this is a workaround as we don't yet have a repeat op that takes custom dim (ggml_repeat_4d)
-        ggml_tensor * repeated = ggml_new_tensor_3d(ctx0, cur->type, n_embd, n_expert_used, n_tokens);
-        repeated = ggml_repeat(ctx0, cur, repeated); // [n_embd, n_expert_used, n_tokens]
+        // repeat cur to [n_embd, n_expert_used, n_tokens]
+        ggml_tensor * repeated = ggml_repeat_4d(ctx0, cur, n_embd, n_expert_used, n_tokens, 1);
         cur = ggml_mul(ctx0, repeated, weights);
         cb(cur, "ffn_moe_weighted", il);
     }
@@ -973,23 +970,6 @@ ggml_tensor * llm_graph_context::build_inp_s_copy() const {
     return cur;
 }
 
-ggml_tensor * llm_graph_context::build_inp_s_mask() const {
-    const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
-
-    auto inp = std::make_unique<llm_graph_input_s_mask>(kv_state);
-
-    const auto n_kv = kv_state->get_n_kv();
-
-    auto & cur = inp->s_mask;
-
-    cur = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, 1, n_kv);
-    ggml_set_input(cur);
-
-    res->add_input(std::move(inp));
-
-    return cur;
-}
-
 ggml_tensor * llm_graph_context::build_inp_cross_embd() const {
     auto inp = std::make_unique<llm_graph_input_cross_embd>(cross);
 
@@ -1442,43 +1422,53 @@ ggml_tensor * llm_graph_context::build_attn(
     return cur;
 }
 
-ggml_tensor * llm_graph_context::build_copy_mask_state(
+ggml_tensor * llm_graph_context::build_recurrent_state(
          ggml_cgraph * gf,
          ggml_tensor * s,
          ggml_tensor * state_copy,
-         ggml_tensor * state_mask,
-             int32_t   n_state,
-             int32_t   n_seqs) const {
+             int32_t   state_size,
+             int32_t   n_seqs,
+                bool   avoid_copies) const {
     const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
 
     const auto n_kv    = kv_state->get_n_kv();
     const auto kv_head = kv_state->get_head();
+    const auto rs_zero = kv_state->get_rs_z();
 
-    ggml_tensor * states = ggml_reshape_2d(ctx0, s, n_state, kv_state->get_size());
+    ggml_tensor * states = ggml_reshape_2d(ctx0, s, state_size, kv_state->get_size());
 
-    // copy states
-    // NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
-    // this shrinks the tensors's ne[1] to n_kv
-    states = ggml_get_rows(ctx0, states, state_copy);
+    // Clear a single state which will then be copied to the other cleared states.
+    // Note that this is a no-op when the view is zero-sized.
+    ggml_tensor * state_zero = ggml_view_1d(ctx0, states, state_size*(rs_zero >= 0), rs_zero*states->nb[1]*(rs_zero >= 0));
+    ggml_build_forward_expand(gf, ggml_scale_inplace(ctx0, state_zero, 0));
 
-    // clear states of sequences which are starting at the beginning of this batch
-    // FIXME: zero-out NANs?
-    states = ggml_mul(ctx0, states, state_mask);
+    ggml_tensor * output_states;
 
-    // copy states which won't be changed further (between n_seqs and n_kv)
+    if (!avoid_copies) {
+        // copy states
+        // NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
+        // {state_size, kv_size} -> {state_size, n_seqs}
+        output_states = ggml_get_rows(ctx0, states, ggml_view_1d(ctx0, state_copy, n_seqs, 0));
+        ggml_build_forward_expand(gf, output_states);
+    } else {
+        // FIXME: make the gathering operation happen before the copy below
+        //        (maybe with an optional lambda function passed as a parameter instead of `avoid_copies`?)
+        output_states = states;
+    }
+
+    // copy extra states which won't be changed further (between n_seqs and n_kv)
+    ggml_tensor * states_extra = ggml_get_rows(ctx0, states, ggml_view_1d(ctx0, state_copy, n_kv - n_seqs, n_seqs*state_copy->nb[0]));
     ggml_build_forward_expand(gf,
         ggml_cpy(ctx0,
-            ggml_view_1d(ctx0, states, n_state*(n_kv - n_seqs), (n_seqs          )*n_state*ggml_element_size(states)),
-            ggml_view_1d(ctx0, s,      n_state*(n_kv - n_seqs), (kv_head + n_seqs)*n_state*ggml_element_size(s))));
+            states_extra,
+            ggml_view_1d(ctx0, s, state_size*(n_kv - n_seqs), (kv_head + n_seqs)*state_size*ggml_element_size(s))));
 
-    // the part of the states that will be used and modified
-    return ggml_view_2d(ctx0, states, n_state, n_seqs, states->nb[1], 0);
+    return output_states;
 }
 
 ggml_tensor * llm_graph_context::build_rwkv_token_shift_load(
          ggml_cgraph * gf,
          ggml_tensor * state_copy,
-         ggml_tensor * state_mask,
   const llama_ubatch & ubatch,
                  int   il) const {
     const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
@@ -1489,8 +1479,8 @@ ggml_tensor * llm_graph_context::build_rwkv_token_shift_load(
 
     ggml_tensor * token_shift_all = kv_state->get_k_l(il);
 
-    ggml_tensor * token_shift = build_copy_mask_state(
-            gf, token_shift_all, state_copy, state_mask,
+    ggml_tensor * token_shift = build_recurrent_state(
+            gf, token_shift_all, state_copy,
             hparams.n_embd_k_s(), n_seqs);
 
     token_shift = ggml_reshape_3d(ctx0, token_shift, hparams.n_embd, token_shift_count, n_seqs);

src/llama-graph.h

@@ -17,7 +17,7 @@ struct ggml_tensor;
 struct llama_ubatch;
 struct llama_cparams;
 
-class llama_memory_state_i;
+struct llama_memory_state_i;
 
 class llama_kv_cache_unified_state;
 class llama_kv_cache_unified_iswa_state;
@@ -36,6 +36,7 @@ enum llm_ffn_op_type {
     LLM_FFN_RELU,
     LLM_FFN_RELU_SQR,
     LLM_FFN_SWIGLU,
+    LLM_FFN_GEGLU,
 };
 
 enum llm_ffn_gate_type {
@@ -199,18 +200,6 @@ public:
     const llama_kv_cache_recurrent_state * kv_state;
 };
 
-class llm_graph_input_s_mask : public llm_graph_input_i {
-public:
-    llm_graph_input_s_mask(const llama_kv_cache_recurrent_state * kv_state) : kv_state(kv_state) {}
-    virtual ~llm_graph_input_s_mask() = default;
-
-    void set_input(const llama_ubatch * ubatch) override;
-
-    ggml_tensor * s_mask; // F32 [1, n_kv]
-
-    const llama_kv_cache_recurrent_state * kv_state;
-};
-
 class llm_graph_input_cross_embd : public llm_graph_input_i {
 public:
     llm_graph_input_cross_embd(
@@ -520,7 +509,6 @@ struct llm_graph_context {
     ggml_tensor * build_inp_mean() const;
     ggml_tensor * build_inp_cls() const;
     ggml_tensor * build_inp_s_copy() const;
-    ggml_tensor * build_inp_s_mask() const;
 
     ggml_tensor * build_inp_cross_embd() const;
     ggml_tensor * build_inp_pos_bucket_enc() const;
@@ -605,18 +593,17 @@ struct llm_graph_context {
     // recurrent
     //
 
-    ggml_tensor * build_copy_mask_state(
+    ggml_tensor * build_recurrent_state(
              ggml_cgraph * gf,
              ggml_tensor * s,
              ggml_tensor * state_copy,
-             ggml_tensor * state_mask,
-                 int32_t   n_state,
-                 int32_t   n_seqs) const;
+                 int32_t   state_size,
+                 int32_t   n_seqs,
+                    bool   avoid_copies = false) const;
 
     ggml_tensor * build_rwkv_token_shift_load(
              ggml_cgraph * gf,
              ggml_tensor * state_copy,
-             ggml_tensor * state_mask,
       const llama_ubatch & ubatch,
                      int   il) const;
 

src/llama-kv-cache-recurrent.cpp

@@ -1,6 +1,7 @@
 #include "llama-kv-cache-recurrent.h"
 
 #include "llama-impl.h"
+#include "llama-io.h"
 #include "llama-batch.h"
 #include "llama-model.h"
 
@@ -116,18 +117,21 @@ llama_kv_cache_recurrent::llama_kv_cache_recurrent(
     }
 }
 
-void llama_kv_cache_recurrent::clear() {
+void llama_kv_cache_recurrent::clear(bool data) {
     for (int32_t i = 0; i < (int32_t) size; ++i) {
         cells[i].pos = -1;
         cells[i].seq_id.clear();
         cells[i].src = -1;
         cells[i].tail = -1;
     }
+
     head = 0;
     used = 0;
 
-    for (auto & buf : bufs) {
-        ggml_backend_buffer_clear(buf.get(), 0);
+    if (data) {
+        for (auto & buf : bufs) {
+            ggml_backend_buffer_clear(buf.get(), 0);
+        }
     }
 }
 
@@ -386,6 +390,13 @@ llama_memory_state_ptr llama_kv_cache_recurrent::init_full() {
     return std::make_unique<llama_kv_cache_recurrent_state>(LLAMA_MEMORY_STATUS_SUCCESS, this);
 }
 
+llama_memory_state_ptr llama_kv_cache_recurrent::init_update(llama_context * lctx, bool optimize) {
+    GGML_UNUSED(lctx);
+    GGML_UNUSED(optimize);
+
+    return std::make_unique<llama_kv_cache_recurrent_state>(LLAMA_MEMORY_STATUS_NO_UPDATE);
+}
+
 bool llama_kv_cache_recurrent::prepare(const std::vector<llama_ubatch> & ubatches) {
     // simply remember the full state because it is very small for this type of cache
     // TODO: optimize
@@ -395,21 +406,12 @@ bool llama_kv_cache_recurrent::prepare(const std::vector<llama_ubatch> & ubatche
 
     bool success = true;
 
-    // TODO: here we have to verify that all ubatches can fit in the cells
-    //       however, the current implementation is broken because it relies on s_copy() and s_mask() to update the cells
-    //         during the compute of each ubatch. to reproduce, uncomment the following loop and run:
-    //
-    //           $ llama-parallel -m ./mamba-130m/ggml-model-f16.gguf -np 5 -ns 8
-    //
-    //       recovery from failures when the batch does not fit in the KV cache will not work correctly until this is fixed
-    //
-    GGML_UNUSED(ubatches);
-    //for (const auto & ubatch : ubatches) {
-    //    if (!find_slot(ubatch)) {
-    //        success = false;
-    //        break;
-    //    }
-    //}
+    for (const auto & ubatch : ubatches) {
+        if (!find_slot(ubatch)) {
+            success = false;
+            break;
+        }
+    }
 
     // restore the original state
     cells = std::move(org_cells);
@@ -419,26 +421,14 @@ bool llama_kv_cache_recurrent::prepare(const std::vector<llama_ubatch> & ubatche
     return success;
 }
 
-bool llama_kv_cache_recurrent::update(llama_context & lctx) {
-    GGML_UNUSED(lctx);
-    // noop
-    return false;
-}
-
-void llama_kv_cache_recurrent::defrag_sched(float thold) {
-    GGML_UNUSED(thold);
-    // noop
-}
-
 bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
-    const uint32_t n_tokens = ubatch.n_tokens;
-    const uint32_t n_seqs   = ubatch.n_seqs;
+    const uint32_t n_seqs = ubatch.n_seqs;
 
     const uint32_t n_seq_tokens = ubatch.n_seq_tokens;
 
     // if we have enough unused cells before the current head ->
     //   better to start searching from the beginning of the cache, hoping to fill it
-    if (head > used + 2*n_tokens) {
+    if (head > used + 2*n_seqs) {
         head = 0;
     }
 
@@ -534,16 +524,16 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
                 empty_cell.src = orig_cell.src;
                 orig_cell.seq_id.erase(seq_id);
                 empty_cell.seq_id.insert(seq_id); // will be overwritten
+                GGML_ASSERT(!orig_cell.is_empty()); // has at least one remaining seq_id
             }
             seq_meta.tail = next_empty_cell;
             // find next empty cell
             if (s + 1 < n_seqs) {
-                next_empty_cell += 1;
                 for (uint32_t i = 0; i < size; ++i) {
+                    next_empty_cell += 1;
                     if (next_empty_cell >= size) { next_empty_cell -= size; }
                     kv_cell & cell = cells[next_empty_cell];
                     if (cell.is_empty()) { break; }
-                    next_empty_cell += 1;
                 }
             }
         }
@@ -553,8 +543,8 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
 
     // gather and re-order
     for (uint32_t s = 0; s < n_seqs; ++s) {
-        int32_t dst_id = s + min;
-        int32_t src_id = cells[ubatch.seq_id[s][0]].tail;
+        const int32_t dst_id = s + min;
+        const int32_t src_id = cells[ubatch.seq_id[s][0]].tail;
         if (dst_id != src_id) {
             kv_cell & dst_cell = cells[dst_id];
             kv_cell & src_cell = cells[src_id];
@@ -563,12 +553,14 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
             std::swap(dst_cell.src, src_cell.src);
             std::swap(dst_cell.seq_id, src_cell.seq_id);
 
-            // swap tails (assuming they NEVER overlap)
-            for (const llama_seq_id seq_id : src_cell.seq_id) {
-                cells[seq_id].tail = src_id;
-            }
-            for (const llama_seq_id seq_id : dst_cell.seq_id) {
-                cells[seq_id].tail = dst_id;
+            // swap tails
+            for (uint32_t i = 0; i < size; ++i) {
+                int32_t & tail = cells[i].tail;
+                if (tail == src_id) {
+                    tail = dst_id;
+                } else if (tail == dst_id) {
+                    tail = src_id;
+                }
             }
         }
     }
@@ -576,7 +568,7 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
     // update the pos of the used seqs
     for (uint32_t s = 0; s < n_seqs; ++s) {
         const llama_pos last_pos = ubatch.pos[n_seq_tokens * s + n_seq_tokens - 1];
-        int32_t cell_id = s + min;
+        const int32_t cell_id = s + min;
         kv_cell & cell = cells[cell_id];
 
         if (cell.pos >= 0 && last_pos != cell.pos + (llama_pos) n_seq_tokens) {
@@ -594,6 +586,38 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
         }
     }
 
+    // Find first cell without src refs, to use as the zero-ed state
+    {
+        // TODO: bake-in src refcounts in the cell metadata
+        std::vector<int32_t> refcounts(size, 0);
+        for (size_t i = 0; i < size; ++i) {
+            const int32_t src = cells[i].src;
+            if (src >= 0) {
+                refcounts[src] += 1;
+            }
+        }
+
+        rs_z = -1;
+        for (int i = min; i <= max; ++i) {
+            if (refcounts[i] == 0) {
+                rs_z = i;
+                break;
+            }
+        }
+
+        for (int i = min; i <= max; ++i) {
+            if (cells[i].src < 0) {
+                GGML_ASSERT(rs_z >= 0);
+                cells[i].src0 = rs_z;
+            } else {
+                // Stage the source ids for all used cells to allow correct seq_* behavior
+                // and still make these values available when setting the inputs
+                cells[i].src0 = cells[i].src;
+            }
+            cells[i].src = i; // avoid moving or clearing twice
+        }
+    }
+
     // allow getting the range of used cells, from head to head + n
     head = min;
     n    = max - min + 1;
@@ -605,47 +629,8 @@ bool llama_kv_cache_recurrent::find_slot(const llama_ubatch & ubatch) {
 }
 
 bool llama_kv_cache_recurrent::get_can_shift() const {
-    return false;
-}
-
-int32_t llama_kv_cache_recurrent::s_copy(int i) const {
-    const uint32_t cell_id = i + head;
-
-    //////////////////////////////////////////////
-    // TODO: this should not mutate the KV cache !
-    kv_cell & cell = const_cast<kv_cell &>(cells[cell_id]);
-
-    // prevent out-of-bound sources
-    if (cell.src < 0 || (uint32_t) cell.src >= size) {
-        cell.src = cell_id;
-    }
-
-    int32_t res = cell.src;
-
-    // TODO: do not mutate the KV cache
-    // ensure copy only happens once
-    if (cell.src != (int32_t) cell_id) {
-        cell.src = cell_id;
-    }
-
-    return res;
-}
-
-float llama_kv_cache_recurrent::s_mask(int i) const {
-    const uint32_t cell_id = i + head;
-
-    //////////////////////////////////////////////
-    // TODO: this should not mutate the KV cache !
-    kv_cell & cell = const_cast<kv_cell &>(cells[cell_id]);
-
-    float res = (float) (cell.src >= 0);
-
-    // only clear once
-    if (cell.src < 0) {
-        cell.src = cell_id;
-    }
-
-    return res;
+    // shifting the pos is trivial for recurrent models
+    return true;
 }
 
 size_t llama_kv_cache_recurrent::total_size() const {
@@ -726,7 +711,7 @@ void llama_kv_cache_recurrent::state_read(llama_io_read_i & io, llama_seq_id seq
 
     if (!res) {
         if (seq_id == -1) {
-            clear();
+            clear(true);
         } else {
             seq_rm(seq_id, -1, -1);
         }
@@ -883,7 +868,7 @@ bool llama_kv_cache_recurrent::state_read_meta(llama_io_read_i & io, uint32_t ce
             return false;
         }
 
-        clear();
+        clear(true);
 
         for (uint32_t i = 0; i < cell_count; ++i) {
             kv_cell & cell = cells[i];
@@ -1111,6 +1096,10 @@ uint32_t llama_kv_cache_recurrent_state::get_head() const {
     return is_full ? 0 : kv->head;
 }
 
+int32_t llama_kv_cache_recurrent_state::get_rs_z() const {
+    return is_full ? 0 : kv->rs_z;
+}
+
 uint32_t llama_kv_cache_recurrent_state::get_size() const {
     return kv->size;
 }
@@ -1124,9 +1113,5 @@ ggml_tensor * llama_kv_cache_recurrent_state::get_v_l(int32_t il) const {
 }
 
 int32_t llama_kv_cache_recurrent_state::s_copy(int i) const {
-    return kv->s_copy(i);
-}
-
-float llama_kv_cache_recurrent_state::s_mask(int i) const {
-    return kv->s_mask(i);
+    return  kv->cells[i + kv->head].src0;
 }

src/llama-kv-cache-recurrent.h

@@ -2,7 +2,7 @@
 
 #include "llama-batch.h"
 #include "llama-graph.h"
-#include "llama-kv-cache.h"
+#include "llama-memory.h"
 
 #include <set>
 #include <vector>
@@ -13,7 +13,7 @@
 
 // TODO: extract the KV cache state used for graph computation into llama_kv_cache_recurrent_state_i
 //       see the implementation of llama_kv_cache_unified_state_i for an example how to do it
-class llama_kv_cache_recurrent : public llama_kv_cache {
+class llama_kv_cache_recurrent : public llama_memory_i {
 public:
     llama_kv_cache_recurrent(
             const llama_model & model,
@@ -29,7 +29,17 @@ public:
     // llama_memory_i
     //
 
-    void clear() override;
+    llama_memory_state_ptr init_batch(
+            const llama_batch & batch,
+            uint32_t n_ubatch,
+            bool embd_pooled,
+            bool logits_all) override;
+
+    llama_memory_state_ptr init_full() override;
+
+    llama_memory_state_ptr init_update(llama_context * lctx, bool optimize) override;
+
+    void clear(bool data) override;
 
     bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) override;
     void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
@@ -40,22 +50,6 @@ public:
     llama_pos seq_pos_min(llama_seq_id seq_id) const override;
     llama_pos seq_pos_max(llama_seq_id seq_id) const override;
 
-    //
-    // llama_kv_cache
-    //
-
-    llama_memory_state_ptr init_batch(
-            const llama_batch & batch,
-            uint32_t n_ubatch,
-            bool embd_pooled,
-            bool logits_all) override;
-
-    llama_memory_state_ptr init_full() override;
-
-    bool update(llama_context & lctx) override;
-
-    void defrag_sched(float thold) override;
-
     bool prepare(const std::vector<llama_ubatch> & ubatches);
 
     // find a contiguous slot of kv cells and emplace the ubatch there
@@ -63,10 +57,6 @@ public:
 
     bool get_can_shift() const override;
 
-    // TODO: temporary methods - they are not really const as they do const_cast<>, fix this
-    int32_t s_copy(int i) const;
-    float   s_mask(int i) const;
-
     // state write/load
 
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
@@ -79,10 +69,14 @@ public:
     // computed before each graph build
     uint32_t n = 0;
 
+    // first zero-ed state
+    int32_t rs_z = -1;
+
     // TODO: optimize for recurrent state needs
     struct kv_cell {
         llama_pos pos  = -1;
-        int32_t   src  = -1; // used to copy states
+        int32_t   src  = -1; // used to know where states should be copied from
+        int32_t   src0 = -1; // like src, but only used when setting the inputs (allowing to copy once)
         int32_t   tail = -1;
 
         std::set<llama_seq_id> seq_id;
@@ -163,13 +157,13 @@ public:
 
     uint32_t get_n_kv() const;
     uint32_t get_head() const;
+    int32_t  get_rs_z() const;
     uint32_t get_size() const;
 
     ggml_tensor * get_k_l(int32_t il) const;
     ggml_tensor * get_v_l(int32_t il) const;
 
     int32_t s_copy(int i) const;
-    float   s_mask(int i) const;
 
 private:
     const llama_memory_status status;

src/llama-kv-cache-unified-iswa.cpp

@@ -52,9 +52,9 @@ llama_kv_cache_unified_iswa::llama_kv_cache_unified_iswa(
             hparams.n_swa, hparams.swa_type);
 }
 
-void llama_kv_cache_unified_iswa::clear() {
-    kv_base->clear();
-    kv_swa ->clear();
+void llama_kv_cache_unified_iswa::clear(bool data) {
+    kv_base->clear(data);
+    kv_swa ->clear(data);
 }
 
 bool llama_kv_cache_unified_iswa::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
@@ -123,26 +123,16 @@ llama_memory_state_ptr llama_kv_cache_unified_iswa::init_batch(const llama_batch
 
     assert(heads_base.size() == heads_swa.size());
 
-    return std::make_unique<llama_kv_cache_unified_iswa_state>(LLAMA_MEMORY_STATUS_SUCCESS,
+    return std::make_unique<llama_kv_cache_unified_iswa_state>(
             this, std::move(sbatch), std::move(heads_base), std::move(heads_swa), std::move(ubatches));
 }
 
 llama_memory_state_ptr llama_kv_cache_unified_iswa::init_full() {
-    return std::make_unique<llama_kv_cache_unified_iswa_state>(LLAMA_MEMORY_STATUS_SUCCESS, this);
+    return std::make_unique<llama_kv_cache_unified_iswa_state>(this);
 }
 
-bool llama_kv_cache_unified_iswa::update(llama_context & lctx) {
-    bool res = false;
-
-    res = res | kv_base->update(lctx);
-    res = res | kv_swa ->update(lctx);
-
-    return res;
-}
-
-void llama_kv_cache_unified_iswa::defrag_sched(float thold) {
-    kv_base->defrag_sched(thold);
-    kv_swa ->defrag_sched(thold);
+llama_memory_state_ptr llama_kv_cache_unified_iswa::init_update(llama_context * lctx, bool optimize) {
+    return std::make_unique<llama_kv_cache_unified_iswa_state>(this, lctx, optimize);
 }
 
 bool llama_kv_cache_unified_iswa::get_can_shift() const {
@@ -174,26 +164,38 @@ llama_kv_cache_unified * llama_kv_cache_unified_iswa::get_swa() const {
 llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(llama_memory_status status) : status(status) {}
 
 llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
-        llama_memory_status status,
-        llama_kv_cache_unified_iswa * kv) : status(status) {
-    state_base.reset(new llama_kv_cache_unified_state(status, kv->get_base()));
-    state_swa .reset(new llama_kv_cache_unified_state(status, kv->get_swa ()));
+        llama_kv_cache_unified_iswa * kv) : status(LLAMA_MEMORY_STATUS_SUCCESS) {
+    state_base = kv->get_base()->init_full();
+    state_swa  = kv->get_swa ()->init_full();
+
+    status = llama_memory_status_combine(state_base->get_status(), state_swa->get_status());
+}
+
+llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
+        llama_kv_cache_unified_iswa * kv,
+        llama_context * lctx,
+        bool optimize) : status(LLAMA_MEMORY_STATUS_SUCCESS) {
+    state_base = kv->get_base()->init_update(lctx, optimize);
+    state_swa  = kv->get_swa ()->init_update(lctx, optimize);
+
+    status = llama_memory_status_combine(state_base->get_status(), state_swa->get_status());
 }
 
 llama_kv_cache_unified_iswa_state::llama_kv_cache_unified_iswa_state(
-        llama_memory_status status,
         llama_kv_cache_unified_iswa * kv,
         llama_sbatch sbatch,
         std::vector<uint32_t> heads_base,
         std::vector<uint32_t> heads_swa,
         std::vector<llama_ubatch> ubatches)
-    : status(status),
-    sbatch(std::move(sbatch)),
-    ubatches(std::move(ubatches)) {
-        // note: here we copy the ubatches. not sure if this is ideal
-        state_base.reset(new llama_kv_cache_unified_state(status, kv->get_base(), {}, std::move(heads_base), this->ubatches));
-        state_swa .reset(new llama_kv_cache_unified_state(status, kv->get_swa (), {}, std::move(heads_swa),  this->ubatches));
-    }
+        : status(LLAMA_MEMORY_STATUS_SUCCESS),
+        sbatch(std::move(sbatch)),
+        ubatches(std::move(ubatches)) {
+    // note: here we copy the ubatches. not sure if this is ideal
+    state_base.reset(new llama_kv_cache_unified_state(kv->get_base(), {}, std::move(heads_base), this->ubatches));
+    state_swa .reset(new llama_kv_cache_unified_state(kv->get_swa (), {}, std::move(heads_swa),  this->ubatches));
+
+    status = llama_memory_status_combine(state_base->get_status(), state_swa->get_status());
+}
 
 llama_kv_cache_unified_iswa_state:: ~llama_kv_cache_unified_iswa_state() = default;
 
@@ -233,17 +235,18 @@ llama_memory_status llama_kv_cache_unified_iswa_state::get_status() const {
 
 const llama_ubatch & llama_kv_cache_unified_iswa_state::get_ubatch() const {
     assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
+
     return ubatches[i_next];
 }
 
 const llama_kv_cache_unified_state * llama_kv_cache_unified_iswa_state::get_base() const {
     assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
 
-    return state_base.get();
+    return static_cast<const llama_kv_cache_unified_state *>(state_base.get());
 }
 
 const llama_kv_cache_unified_state * llama_kv_cache_unified_iswa_state::get_swa()  const {
     assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
 
-    return state_swa.get();
+    return static_cast<const llama_kv_cache_unified_state *>(state_swa.get());
 }

src/llama-kv-cache-unified-iswa.h

@@ -11,7 +11,7 @@
 // utilizes two instances of llama_kv_cache_unified
 //   the first instance is for the non-SWA layers of the model and the second instance is for the SWA layers
 
-class llama_kv_cache_unified_iswa : public llama_kv_cache {
+class llama_kv_cache_unified_iswa : public llama_memory_i {
 public:
     llama_kv_cache_unified_iswa(
             const llama_model & model,
@@ -31,7 +31,19 @@ public:
     // llama_memory_i
     //
 
-    void clear() override;
+    llama_memory_state_ptr init_batch(
+            const llama_batch & batch,
+            uint32_t n_ubatch,
+            bool embd_pooled,
+            bool logits_all) override;
+
+    llama_memory_state_ptr init_full() override;
+
+    llama_memory_state_ptr init_update(llama_context * lctx, bool optimize) override;
+
+    bool get_can_shift() const override;
+
+    void clear(bool data) override;
 
     bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) override;
     void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
@@ -42,24 +54,6 @@ public:
     llama_pos seq_pos_min(llama_seq_id seq_id) const override;
     llama_pos seq_pos_max(llama_seq_id seq_id) const override;
 
-    //
-    // llama_kv_cache
-    //
-
-    llama_memory_state_ptr init_batch(
-            const llama_batch & batch,
-            uint32_t n_ubatch,
-            bool embd_pooled,
-            bool logits_all) override;
-
-    llama_memory_state_ptr init_full() override;
-
-    bool update(llama_context & lctx) override;
-
-    void defrag_sched(float thold) override;
-
-    bool get_can_shift() const override;
-
     // state write/load
 
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
@@ -86,12 +80,16 @@ public:
 
     // used to create a full-cache state
     llama_kv_cache_unified_iswa_state(
-            llama_memory_status status,
             llama_kv_cache_unified_iswa * kv);
 
+    // used to create an update state
+    llama_kv_cache_unified_iswa_state(
+            llama_kv_cache_unified_iswa * kv,
+            llama_context * lctx,
+            bool optimize);
+
     // used to create a state from a batch
     llama_kv_cache_unified_iswa_state(
-            llama_memory_status status,
             llama_kv_cache_unified_iswa * kv,
             llama_sbatch sbatch,
             std::vector<uint32_t> heads_base,
@@ -120,7 +118,7 @@ public:
     const llama_kv_cache_unified_state * get_swa()  const;
 
 private:
-    const llama_memory_status status;
+    llama_memory_status status;
 
     //llama_kv_cache_unified_iswa * kv;
 
@@ -131,6 +129,6 @@ private:
 
     std::vector<llama_ubatch> ubatches;
 
-    std::unique_ptr<llama_kv_cache_unified_state> state_base;
-    std::unique_ptr<llama_kv_cache_unified_state> state_swa;
+    llama_memory_state_ptr state_base;
+    llama_memory_state_ptr state_swa;
 };

src/llama-kv-cache-unified.cpp

@@ -1,6 +1,7 @@
 #include "llama-kv-cache-unified.h"
 
 #include "llama-impl.h"
+#include "llama-io.h"
 #include "llama-model.h"
 #include "llama-context.h"
 
@@ -128,13 +129,15 @@ llama_kv_cache_unified::llama_kv_cache_unified(
     }
 }
 
-void llama_kv_cache_unified::clear() {
+void llama_kv_cache_unified::clear(bool data) {
     cells.reset();
 
     head = 0;
 
-    for (auto & buf : bufs) {
-        ggml_backend_buffer_clear(buf.get(), 0);
+    if (data) {
+        for (auto & buf : bufs) {
+            ggml_backend_buffer_clear(buf.get(), 0);
+        }
     }
 }
 
@@ -149,12 +152,27 @@ bool llama_kv_cache_unified::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos
         p1 = std::numeric_limits<llama_pos>::max();
     }
 
-    for (uint32_t i = 0; i < cells.size(); ++i) {
-        if (!cells.pos_in(i, p0, p1)) {
-            continue;
-        }
+    if (seq_id >= 0) {
+        for (uint32_t i = 0; i < cells.size(); ++i) {
+            if (!cells.pos_in(i, p0, p1)) {
+                continue;
+            }
+
+            if (cells.seq_has(i, seq_id) && cells.seq_rm(i, seq_id)) {
+                if (new_head == cells.size()) {
+                    new_head = i;
+                }
+            }
+        }
+    } else {
+        // match any sequence
+        for (uint32_t i = 0; i < cells.size(); ++i) {
+            if (!cells.pos_in(i, p0, p1)) {
+                continue;
+            }
+
+            cells.rm(i);
 
-        if (cells.seq_has(i, seq_id) && cells.seq_rm(i, seq_id)) {
             if (new_head == cells.size()) {
                 new_head = i;
             }
@@ -305,16 +323,49 @@ llama_memory_state_ptr llama_kv_cache_unified::init_batch(
         return std::make_unique<llama_kv_cache_unified_state>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
     }
 
-    return std::make_unique<llama_kv_cache_unified_state>(LLAMA_MEMORY_STATUS_SUCCESS,
+    return std::make_unique<llama_kv_cache_unified_state>(
             this, std::move(sbatch), std::move(heads), std::move(ubatches));
 }
 
 llama_memory_state_ptr llama_kv_cache_unified::init_full() {
-    return std::make_unique<llama_kv_cache_unified_state>(LLAMA_MEMORY_STATUS_SUCCESS, this);
+    return std::make_unique<llama_kv_cache_unified_state>(this);
 }
 
-std::vector<uint32_t> llama_kv_cache_unified::prepare(const std::vector<llama_ubatch> & ubatches) {
-    std::vector<uint32_t> res;
+llama_memory_state_ptr llama_kv_cache_unified::init_update(llama_context * lctx, bool optimize) {
+    bool do_shift = get_has_shift();
+
+    defrag_info dinfo;
+
+    // see if we need to defrag
+    {
+        bool do_defrag = optimize;
+
+        const auto thold = lctx->get_cparams().defrag_thold;
+
+        if (!do_defrag && thold > 0.0f) {
+            const auto n_kv = cells.used_max_p1();
+
+            // - do not defrag small contexts (i.e. < 2048 tokens)
+            // - count the padding towards the number of used tokens
+            const float fragmentation = n_kv >= 2048 ? std::max(0.0f, 1.0f - (float(cells.get_used() + n_pad)/n_kv)) : 0.0f;
+
+            if (fragmentation > thold) {
+                LLAMA_LOG_DEBUG("%s: fragmentation: %.2f - requesting defrag\n", __func__, fragmentation);
+
+                do_defrag = true;
+            }
+        }
+
+        if (do_defrag) {
+            dinfo = defrag_prepare(lctx->graph_max_nodes());
+        }
+    }
+
+    return std::make_unique<llama_kv_cache_unified_state>(this, lctx, do_shift, std::move(dinfo));
+}
+
+llama_kv_cache_unified::ubatch_heads llama_kv_cache_unified::prepare(const std::vector<llama_ubatch> & ubatches) {
+    llama_kv_cache_unified::ubatch_heads res;
 
     struct state {
         uint32_t head_old; // old position of the head, before placing the ubatch
@@ -359,12 +410,12 @@ std::vector<uint32_t> llama_kv_cache_unified::prepare(const std::vector<llama_ub
     return res;
 }
 
-bool llama_kv_cache_unified::update(llama_context & lctx) {
+bool llama_kv_cache_unified::update(llama_context * lctx, bool do_shift, const defrag_info & dinfo) {
     bool updated = false;
 
-    auto * sched = lctx.get_sched();
+    auto * sched = lctx->get_sched();
 
-    if (cells.get_has_shift()) {
+    if (do_shift) {
         if (!get_can_shift()) {
             GGML_ABORT("The current KV cache / model configuration does not support K-shift");
         }
@@ -375,9 +426,9 @@ bool llama_kv_cache_unified::update(llama_context & lctx) {
         if (hparams.rope_type != LLAMA_ROPE_TYPE_NONE) {
             ggml_backend_sched_reset(sched);
 
-            auto * gf = lctx.graph_init();
+            auto * gf = lctx->graph_init();
 
-            auto res = build_graph_shift(lctx.get_cparams(), lctx.get_ctx_compute(), gf);
+            auto res = build_graph_shift(lctx->get_cparams(), lctx->get_ctx_compute(), gf);
             if (!res) {
                 LLAMA_LOG_ERROR("%s: failed to build graph for K-shift\n", __func__);
                 return updated;
@@ -390,7 +441,7 @@ bool llama_kv_cache_unified::update(llama_context & lctx) {
 
             res->set_inputs(nullptr);
 
-            if (lctx.graph_compute(gf, false) != GGML_STATUS_SUCCESS) {
+            if (lctx->graph_compute(gf, false) != GGML_STATUS_SUCCESS) {
                 LLAMA_LOG_ERROR("%s: failed to compute K-shift\n", __func__);
                 return updated;
             }
@@ -401,56 +452,55 @@ bool llama_kv_cache_unified::update(llama_context & lctx) {
         cells.reset_shift();
     }
 
-    if (do_defrag) {
+    if (!dinfo.empty()) {
         LLAMA_LOG_DEBUG("%s: defragmenting KV cache\n", __func__);
 
-        if (defrag_prepare(lctx.graph_max_nodes())) {
-            ggml_backend_sched_reset(sched);
+        // apply moves:
+        {
+            const auto n_kv = dinfo.ids.size();
 
-            auto * gf = lctx.graph_init();
+            for (uint32_t i = 0; i < n_kv; ++i) {
+                assert(dinfo.ids[i] <= n_kv);
 
-            auto res = build_graph_defrag(lctx.get_cparams(), lctx.get_ctx_compute(), gf);
-            if (!res) {
-                LLAMA_LOG_ERROR("%s: failed to build graph for defrag\n", __func__);
-                return updated;
+                if (dinfo.ids[i] == n_kv) {
+                    continue;
+                }
+
+                cells.mv(i, dinfo.ids[i]);
             }
 
-            if (!ggml_backend_sched_alloc_graph(sched, gf)) {
-                LLAMA_LOG_ERROR("%s: failed to allocate compute graph for defrag\n", __func__);
-                return updated;
-            }
-
-            res->set_inputs(nullptr);
-
-            if (lctx.graph_compute(gf, false) != GGML_STATUS_SUCCESS) {
-                LLAMA_LOG_ERROR("%s: failed to compute defrag\n", __func__);
-                return updated;
-            }
-
-            updated = true;
+            // reset the head so we can find the first free slot during the next ubatch
+            head = 0;
         }
 
-        do_defrag = false;
+        ggml_backend_sched_reset(sched);
+
+        auto * gf = lctx->graph_init();
+
+        auto res = build_graph_defrag(lctx->get_cparams(), lctx->get_ctx_compute(), gf, dinfo);
+        if (!res) {
+            LLAMA_LOG_ERROR("%s: failed to build graph for defrag\n", __func__);
+            return updated;
+        }
+
+        if (!ggml_backend_sched_alloc_graph(sched, gf)) {
+            LLAMA_LOG_ERROR("%s: failed to allocate compute graph for defrag\n", __func__);
+            return updated;
+        }
+
+        res->set_inputs(nullptr);
+
+        if (lctx->graph_compute(gf, false) != GGML_STATUS_SUCCESS) {
+            LLAMA_LOG_ERROR("%s: failed to compute defrag\n", __func__);
+            return updated;
+        }
+
+        updated = true;
     }
 
     return updated;
 }
 
-void llama_kv_cache_unified::defrag_sched(float thold) {
-    const auto n_kv = cells.used_max_p1();
-
-    // - do not defrag small contexts (i.e. < 2048 tokens)
-    // - count the padding towards the number of used tokens
-    const float fragmentation = n_kv >= 2048 ? std::max(0.0f, 1.0f - (float(cells.get_used() + n_pad)/n_kv)) : 0.0f;
-
-    // queue defragmentation for next llama_kv_cache_update
-    if (fragmentation > thold) {
-        LLAMA_LOG_DEBUG("%s: fragmentation: %.2f - requesting defrag\n", __func__, fragmentation);
-
-        do_defrag = true;
-    }
-}
-
 int32_t llama_kv_cache_unified::find_slot(const llama_ubatch & ubatch) const {
     const uint32_t n_tokens = ubatch.n_tokens;
 
@@ -462,8 +512,6 @@ int32_t llama_kv_cache_unified::find_slot(const llama_ubatch & ubatch) const {
         head_cur = 0;
     }
 
-    // otherwise, one cell per token.
-
     if (n_tokens > cells.size()) {
         LLAMA_LOG_ERROR("%s: n_tokens = %d > size = %u\n", __func__, n_tokens, cells.size());
         return -1;
@@ -597,6 +645,10 @@ uint32_t llama_kv_cache_unified::get_size() const {
     return cells.size();
 }
 
+bool llama_kv_cache_unified::get_has_shift() const {
+    return cells.get_has_shift();
+}
+
 uint32_t llama_kv_cache_unified::get_n_kv() const {
     return std::min(cells.size(), std::max(n_pad, GGML_PAD(cells.used_max_p1(), n_pad)));
 }
@@ -926,12 +978,13 @@ llm_graph_result_ptr llama_kv_cache_unified::build_graph_shift(
 }
 
 llm_graph_result_ptr llama_kv_cache_unified::build_graph_defrag(
-        const llama_cparams & cparams,
-               ggml_context * ctx,
-                ggml_cgraph * gf) const {
+                const llama_cparams & cparams,
+                       ggml_context * ctx,
+                        ggml_cgraph * gf,
+                  const defrag_info & dinfo) const {
     auto res = std::make_unique<llm_graph_result>();
 
-    const auto & ids = defrag_info.ids;
+    const auto & ids = dinfo.ids;
 
 #if 0
     // CPU defrag
@@ -1072,7 +1125,7 @@ llm_graph_result_ptr llama_kv_cache_unified::build_graph_defrag(
     return res;
 }
 
-bool llama_kv_cache_unified::defrag_prepare(int32_t n_max_nodes) {
+llama_kv_cache_unified::defrag_info llama_kv_cache_unified::defrag_prepare(int32_t n_max_nodes) const {
     const uint32_t n_layer = layers.size();
 
     const uint32_t n_kv   = cells.used_max_p1();
@@ -1093,14 +1146,9 @@ bool llama_kv_cache_unified::defrag_prepare(int32_t n_max_nodes) {
     const uint32_t max_moves = (n_max_nodes - 2*n_layer)/(6*n_layer);
 
     // determine which KV cells to move where
-    //
-    //  cell i moves to ids[i]
-    //
-    //  if ids[i] == i || ids[i] == n_kv, then cell i is not moved
-    //
-    auto & ids = defrag_info.ids;
+    defrag_info res;
+    auto & ids = res.ids;
 
-    ids.clear();
     ids.resize(n_kv, n_kv);
 
     for (uint32_t i0 = 0; i0 < n_used; ++i0) {
@@ -1164,11 +1212,6 @@ bool llama_kv_cache_unified::defrag_prepare(int32_t n_max_nodes) {
             // this cell goes to (i0 + nf)
             ids[i1] = i0 + nf;
 
-            // move the cell meta data
-            cells.mv(i1, i0 + nf);
-
-            head = n_used;
-
             if (!cont) {
                 n_moves++;
                 cont = true;
@@ -1191,14 +1234,14 @@ bool llama_kv_cache_unified::defrag_prepare(int32_t n_max_nodes) {
     }
 
     if (n_moves == 0) {
-        return false;
+        return {};
     }
 
     LLAMA_LOG_DEBUG("%s: (tmp log) KV defrag cell moves: %u\n", __func__, n_moves);
 
     LLAMA_LOG_DEBUG("%s: expected gf nodes: %u\n", __func__, 6*n_moves*n_layer);
 
-    return true;
+    return res;
 }
 
 bool llama_kv_cache_unified::is_masked_swa(llama_pos p0, llama_pos p1) const {
@@ -1276,7 +1319,7 @@ void llama_kv_cache_unified::state_read(llama_io_read_i & io, llama_seq_id seq_i
 
     if (!res) {
         if (seq_id == -1) {
-            clear();
+            clear(true);
         } else {
             seq_rm(seq_id, -1, -1);
         }
@@ -1457,7 +1500,7 @@ bool llama_kv_cache_unified::state_read_meta(llama_io_read_i & io, uint32_t cell
             return false;
         }
 
-        clear();
+        clear(true);
 
         for (uint32_t i = 0; i < cell_count; ++i) {
             llama_pos pos;
@@ -1621,24 +1664,27 @@ bool llama_kv_cache_unified::state_read_data(llama_io_read_i & io, uint32_t cell
 llama_kv_cache_unified_state::llama_kv_cache_unified_state(llama_memory_status status) : status(status) {}
 
 llama_kv_cache_unified_state::llama_kv_cache_unified_state(
-            llama_memory_status status,
-            llama_kv_cache_unified * kv) : status(status), kv(kv) {
-        n_kv = kv->get_size();
-        head = 0;
-    }
+        llama_kv_cache_unified * kv) : status(LLAMA_MEMORY_STATUS_SUCCESS), kv(kv) {
+    n_kv = kv->get_size();
+    head = 0;
+}
 
 llama_kv_cache_unified_state::llama_kv_cache_unified_state(
-            llama_memory_status status,
-            llama_kv_cache_unified * kv,
-            llama_sbatch sbatch,
-            std::vector<uint32_t> heads,
-            std::vector<llama_ubatch> ubatches)
-            : status(status),
-              kv(kv),
-              sbatch(std::move(sbatch)),
-              heads(std::move(heads)),
-              ubatches(std::move(ubatches)) {
+        llama_kv_cache_unified * kv,
+        llama_context * lctx,
+        bool do_shift,
+        defrag_info dinfo) : status(LLAMA_MEMORY_STATUS_SUCCESS), kv(kv), lctx(lctx), do_shift(do_shift), dinfo(std::move(dinfo)) {
+    if (!do_shift && dinfo.empty()) {
+        status = LLAMA_MEMORY_STATUS_NO_UPDATE;
     }
+}
+
+llama_kv_cache_unified_state::llama_kv_cache_unified_state(
+        llama_kv_cache_unified * kv,
+        llama_sbatch sbatch,
+        llama_kv_cache_unified::ubatch_heads heads,
+        std::vector<llama_ubatch> ubatches) : status(LLAMA_MEMORY_STATUS_SUCCESS), kv(kv), sbatch(std::move(sbatch)), heads(std::move(heads)), ubatches(std::move(ubatches)) {
+}
 
 llama_kv_cache_unified_state::~llama_kv_cache_unified_state() = default;
 
@@ -1655,6 +1701,13 @@ bool llama_kv_cache_unified_state::next() {
 bool llama_kv_cache_unified_state::apply() {
     assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
 
+    // no ubatches -> this is a KV cache update
+    if (ubatches.empty()) {
+        kv->update(lctx, do_shift, dinfo);
+
+        return true;
+    }
+
     kv->apply_ubatch(heads[i_next], ubatches[i_next]);
 
     n_kv = kv->get_n_kv();

src/llama-kv-cache-unified.h

@@ -2,8 +2,8 @@
 
 #include "llama-batch.h"
 #include "llama-graph.h"
-#include "llama-kv-cache.h"
 #include "llama-kv-cells.h"
+#include "llama-memory.h"
 
 #include <unordered_map>
 #include <vector>
@@ -17,13 +17,26 @@ struct llama_context;
 // llama_kv_cache_unified
 //
 
-class llama_kv_cache_unified : public llama_kv_cache {
+class llama_kv_cache_unified : public llama_memory_i {
 public:
     static uint32_t get_padding(const llama_cparams & cparams);
 
     // this callback is used to filter out layers that should not be included in the cache
     using layer_filter_cb = std::function<bool(int32_t il)>;
 
+    using ubatch_heads = std::vector<uint32_t>;
+
+    struct defrag_info {
+        bool empty() const {
+            return ids.empty();
+        }
+
+        // contains information about which cell moves where:
+        //  - cell i moves to ids[i]
+        //  - if ids[i] == i || ids[i] == ids.size(), then cell i is not moved
+        std::vector<uint32_t> ids;
+    };
+
     llama_kv_cache_unified(
             const llama_model &  model,
               layer_filter_cb && filter,
@@ -43,7 +56,19 @@ public:
     // llama_memory_i
     //
 
-    void clear() override;
+    llama_memory_state_ptr init_batch(
+            const llama_batch & batch,
+            uint32_t n_ubatch,
+            bool embd_pooled,
+            bool logits_all) override;
+
+    llama_memory_state_ptr init_full() override;
+
+    llama_memory_state_ptr init_update(llama_context * lctx, bool optimize) override;
+
+    bool get_can_shift() const override;
+
+    void clear(bool data) override;
 
     bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) override;
     void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
@@ -54,24 +79,6 @@ public:
     llama_pos seq_pos_min(llama_seq_id seq_id) const override;
     llama_pos seq_pos_max(llama_seq_id seq_id) const override;
 
-    //
-    // llama_kv_cache
-    //
-
-    llama_memory_state_ptr init_batch(
-            const llama_batch & batch,
-            uint32_t n_ubatch,
-            bool embd_pooled,
-            bool logits_all) override;
-
-    llama_memory_state_ptr init_full() override;
-
-    bool update(llama_context & lctx) override;
-
-    void defrag_sched(float thold) override;
-
-    bool get_can_shift() const override;
-
     // state write/load
 
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
@@ -83,6 +90,8 @@ public:
 
     uint32_t get_size() const;
 
+    bool get_has_shift() const;
+
     //
     // graph_build API
     //
@@ -103,7 +112,9 @@ public:
 
     // find places for the provided ubatches in the cache, returns the head locations
     // return empty vector on failure
-    std::vector<uint32_t> prepare(const std::vector<llama_ubatch> & ubatches);
+    ubatch_heads prepare(const std::vector<llama_ubatch> & ubatches);
+
+    bool update(llama_context * lctx, bool do_shift, const defrag_info & dinfo);
 
     // return the cell position where we can insert the ubatch
     // return -1 on failure to find a contiguous slot of kv cells
@@ -133,8 +144,7 @@ private:
         ggml_tensor * v;
     };
 
-    bool do_defrag = false;
-    bool v_trans   = true;  // the value tensor is transposed
+    bool v_trans = true;  // the value tensor is transposed
 
     // the current index from where we start searching for a free slot in the ring buffer of KV cells (see find_slot())
     // note: this is not part of the KV state and it's only used to speed-up the find_slot() method
@@ -160,13 +170,8 @@ private:
     // model layer id -> KV cache layer id
     std::unordered_map<int32_t, int32_t> map_layer_ids;
 
-    // defrag
-    struct {
-        std::vector<uint32_t> ids;
-    } defrag_info;
-
-    // return true if cells have been moved
-    bool defrag_prepare(int32_t n_max_nodes);
+    // return non-empty vector if cells have been moved
+    defrag_info defrag_prepare(int32_t n_max_nodes) const;
 
     size_t total_size() const;
 
@@ -192,7 +197,8 @@ private:
     llm_graph_result_ptr build_graph_defrag(
             const llama_cparams & cparams,
                    ggml_context * ctx,
-                    ggml_cgraph * gf) const;
+                    ggml_cgraph * gf,
+              const defrag_info & dinfo) const;
 
     void state_write_meta(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges, llama_seq_id seq_id = -1) const;
     void state_write_data(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges) const;
@@ -203,20 +209,29 @@ private:
 
 class llama_kv_cache_unified_state : public llama_memory_state_i {
 public:
+    // some shorthands
+    using ubatch_heads = llama_kv_cache_unified::ubatch_heads;
+    using defrag_info  = llama_kv_cache_unified::defrag_info;
+
     // used for errors
     llama_kv_cache_unified_state(llama_memory_status status);
 
     // used to create a full-cache state
     llama_kv_cache_unified_state(
-            llama_memory_status status,
             llama_kv_cache_unified * kv);
 
-    // used to create a state from a batch
+    // used to create an update state
+    llama_kv_cache_unified_state(
+            llama_kv_cache_unified * kv,
+            llama_context * lctx,
+            bool do_shift,
+            defrag_info dinfo);
+
+    // used to create a decode state from a batch
     llama_kv_cache_unified_state(
-            llama_memory_status status,
             llama_kv_cache_unified * kv,
             llama_sbatch sbatch,
-            std::vector<uint32_t> heads,
+            ubatch_heads heads,
             std::vector<llama_ubatch> ubatches);
 
     virtual ~llama_kv_cache_unified_state();
@@ -253,16 +268,30 @@ public:
     void set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const;
 
 private:
-    const llama_memory_status status;
+    llama_memory_status status;
 
     llama_kv_cache_unified * kv;
+    llama_context * lctx;
+
+    //
+    // update state
+    //
+
+    bool do_shift = false;
+
+    defrag_info dinfo;
+
+    //
+    // batch processing state
+    //
 
     llama_sbatch sbatch;
 
     // the index of the next ubatch to process
     size_t i_next = 0;
 
-    std::vector<uint32_t> heads;
+    ubatch_heads heads;
+
     std::vector<llama_ubatch> ubatches;
 
     //

src/llama-kv-cache.cpp

@@ -1 +0,0 @@
-#include "llama-kv-cache.h"

src/llama-kv-cache.h

@@ -1,44 +0,0 @@
-#pragma once
-
-#include "llama.h"
-#include "llama-io.h"
-#include "llama-memory.h"
-
-struct llama_kv_cache : public llama_memory_i {
-    virtual ~llama_kv_cache() = default;
-
-    // split the input batch into a set of ubatches and verify that they can fit into the cache
-    // return a state object containing the ubatches and KV cache state required to process them
-    // check the llama_memory_state_i::get_status() for the result
-    virtual llama_memory_state_ptr init_batch(
-            const llama_batch & batch,
-            uint32_t n_ubatch,
-            bool embd_pooled,
-            bool logits_all) = 0;
-
-    // simulate full cache, used for allocating worst-case compute buffers
-    virtual llama_memory_state_ptr init_full() = 0;
-
-    // process any pending defrag/shift/etc. operations
-    // optionally call once before processing a new batch
-    // return true if any operations were performed
-    virtual bool update(llama_context & lctx) = 0;
-
-    // schedule a defrag if the fragmentation threshold is exceeded. otherwise, do nothing
-    // TODO: change to
-    //   llama_memory_state_ptr init_defrag(float thold) = 0;
-    //
-    virtual void defrag_sched(float thold) = 0;
-
-    // getters
-    virtual bool get_can_shift() const = 0;
-
-    bool get_can_edit() const override { return get_can_shift(); }
-
-    //
-    // state write/read
-    //
-
-    virtual void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const = 0;
-    virtual void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1) = 0;
-};

src/llama-memory.cpp

@@ -1 +1,42 @@
 #include "llama-memory.h"
+
+llama_memory_status llama_memory_status_combine(llama_memory_status s0, llama_memory_status s1) {
+    bool has_update = false;
+
+    switch (s0) {
+        case LLAMA_MEMORY_STATUS_SUCCESS:
+            {
+                has_update = true;
+                break;
+            }
+        case LLAMA_MEMORY_STATUS_NO_UPDATE:
+            {
+                break;
+            }
+        case LLAMA_MEMORY_STATUS_FAILED_PREPARE:
+        case LLAMA_MEMORY_STATUS_FAILED_COMPUTE:
+            {
+                return s0;
+            }
+    }
+
+    switch (s1) {
+        case LLAMA_MEMORY_STATUS_SUCCESS:
+            {
+                has_update = true;
+                break;
+            }
+        case LLAMA_MEMORY_STATUS_NO_UPDATE:
+            {
+                break;
+            }
+        case LLAMA_MEMORY_STATUS_FAILED_PREPARE:
+        case LLAMA_MEMORY_STATUS_FAILED_COMPUTE:
+            {
+                return s1;
+            }
+    }
+
+    // if either status has an update, then the combined status has an update
+    return has_update ? LLAMA_MEMORY_STATUS_SUCCESS : LLAMA_MEMORY_STATUS_NO_UPDATE;
+}

src/llama-memory.h

@@ -7,6 +7,9 @@
 
 struct llama_ubatch;
 
+class llama_io_write_i;
+class llama_io_read_i;
+
 struct llama_memory_params {
     // kv cache
     ggml_type type_k;
@@ -16,32 +19,17 @@ struct llama_memory_params {
     bool swa_full;
 };
 
-// general concept of LLM memory
-// the KV cache is a type of LLM memory, but there can be other types
-class llama_memory_i {
-public:
-    virtual ~llama_memory_i() = default;
-
-    virtual void clear() = 0;
-
-    virtual bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) = 0;
-    virtual void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) = 0;
-    virtual void seq_keep(llama_seq_id seq_id) = 0;
-    virtual void seq_add (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, llama_pos shift) = 0;
-    virtual void seq_div (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, int d) = 0;
-
-    virtual llama_pos seq_pos_min(llama_seq_id seq_id) const = 0;
-    virtual llama_pos seq_pos_max(llama_seq_id seq_id) const = 0;
-
-    virtual bool get_can_edit() const = 0;
-};
-
 enum llama_memory_status {
     LLAMA_MEMORY_STATUS_SUCCESS = 0,
+    LLAMA_MEMORY_STATUS_NO_UPDATE,
     LLAMA_MEMORY_STATUS_FAILED_PREPARE,
     LLAMA_MEMORY_STATUS_FAILED_COMPUTE,
 };
 
+// helper function for combining the status of two memory states
+// useful for implementing hybrid memory types (e.g. iSWA)
+llama_memory_status llama_memory_status_combine(llama_memory_status s0, llama_memory_status s1);
+
 // the interface for managing the memory state during batch processing
 // this interface is implemented per memory type. see:
 //   - llama_kv_cache_unified_state
@@ -51,8 +39,7 @@ enum llama_memory_status {
 // the only method that can mutate the memory and the memory state is llama_memory_i::apply()
 //
 // TODO: rename to llama_memory_context_i ?
-class llama_memory_state_i {
-public:
+struct llama_memory_state_i {
     virtual ~llama_memory_state_i() = default;
 
     // consume the current ubatch from the state and proceed to the next one
@@ -69,8 +56,63 @@ public:
     // get the current ubatch
     virtual const llama_ubatch & get_ubatch() const = 0;
 
-    // get the status of the memory state
+    // get the status of the memory state - used for error handling and checking if any updates would be applied
     virtual llama_memory_status get_status() const = 0;
 };
 
 using llama_memory_state_ptr = std::unique_ptr<llama_memory_state_i>;
+
+// general concept of LLM memory
+// the KV cache is a type of LLM memory, but there can be other types
+struct llama_memory_i {
+    virtual ~llama_memory_i() = default;
+
+    // split the input batch into a set of ubatches and verify that they can fit into the cache
+    // return a state object containing the ubatches and KV cache state required to process them
+    // check the llama_memory_state_i::get_status() for the result
+    virtual llama_memory_state_ptr init_batch(
+            const llama_batch & batch,
+            uint32_t n_ubatch,
+            bool embd_pooled,
+            bool logits_all) = 0;
+
+    // simulate full cache, used for allocating worst-case compute buffers
+    virtual llama_memory_state_ptr init_full() = 0;
+
+    // prepare for any pending memory updates, such as shifts, defrags, etc.
+    // status == LLAMA_MEMORY_STATUS_NO_UPDATE if there is nothing to update
+    virtual llama_memory_state_ptr init_update(llama_context * lctx, bool optimize) = 0;
+
+    // getters
+    virtual bool get_can_shift() const = 0;
+
+    //
+    // ops
+    //
+
+    // if data == true, the data buffers will also be cleared together with the metadata
+    virtual void clear(bool data) = 0;
+
+    virtual bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) = 0;
+    virtual void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) = 0;
+    virtual void seq_keep(llama_seq_id seq_id) = 0;
+    virtual void seq_add (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, llama_pos shift) = 0;
+    virtual void seq_div (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, int d) = 0;
+
+    virtual llama_pos seq_pos_min(llama_seq_id seq_id) const = 0;
+    virtual llama_pos seq_pos_max(llama_seq_id seq_id) const = 0;
+
+    //
+    // state write/read
+    //
+
+    virtual void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const = 0;
+    virtual void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1) = 0;
+};
+
+using llama_memory_ptr = std::unique_ptr<llama_memory_i>;
+
+// TODO: temporary until the llama_kv_cache is removed from the public API
+struct llama_kv_cache : public llama_memory_i {
+    virtual ~llama_kv_cache() = default;
+};

src/llama-mmap.cpp

@@ -401,7 +401,7 @@ struct llama_mmap::impl {
                 }
             }
 #else
-            throw std::runtime_error("PrefetchVirtualMemory unavailable");
+            LLAMA_LOG_DEBUG("skipping PrefetchVirtualMemory because _WIN32_WINNT < 0x602\n");
 #endif
         }
     }