Merge fc6abde7aa into c35e586ea5

* mtgpu: add mp_21 support

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

* mtgpu: disable flash attention on qy1 (MTT S80); disable q3_k and mul_mat_batched_cublas

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

* mtgpu: enable unified memory

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

* mtgpu: map cublasOperation_t to mublasOperation_t (sync code to latest)

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

---------

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

.github/workflows/server-convert-and-infer.yml

@@ -0,0 +1,77 @@
+# Server build and tests conversion from safetensors and inference
+name: LoRA adapters conversion and inference
+
+on:
+  workflow_dispatch: # allows manual triggering
+    inputs:
+      sha:
+        description: 'Commit SHA1 to build'
+        required: false
+        type: string
+      slow_tests:
+        description: 'Run slow tests'
+        required: true
+        type: boolean
+  push:
+    branches:
+      - master
+    paths: ['.github/workflows/server.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', 'examples/server/**.*']
+  pull_request:
+    types: [opened, synchronize, reopened]
+    paths: ['.github/workflows/server.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', 'examples/server/**.*']
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  server:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          sudo apt-get -y install \
+            build-essential \
+            git \
+            cmake \
+            curl \
+          curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | bash
+          sudo apt-get install -y git-lfs
+
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
+
+      - name: Python setup
+        id: setup_python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+
+      - name: Tests dependencies
+        id: test_dependencies
+        run: |
+          pip install -r requirements/requirements-all.txt
+
+      - name: Build
+        id: cmake_build
+        if: ${{ matrix.sanitizer != 'THREAD' }}
+        run: |
+          make llama-cli llama-export-lora \
+              -DGGML_NATIVE=OFF \
+              -DLLAMA_BUILD_SERVER=ON \
+              -DLLAMA_CURL=ON \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
+              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
+
+      - name: Lora convert and inference tests
+        id: test_lora_conversion_inference
+        if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
+        run: ./tests/test-lora-conversion-inference.sh

Makefile

@@ -611,7 +611,7 @@ ifdef GGML_CUDA
 
 		MK_CPPFLAGS  += -DGGML_USE_CUDA -I$(CUDA_PATH)/include
 		MK_LDFLAGS   += -lmusa -lmublas -lmusart -lpthread -ldl -lrt -L$(CUDA_PATH)/lib -L/usr/lib64
-		MK_NVCCFLAGS += -x musa -mtgpu --cuda-gpu-arch=mp_22
+		MK_NVCCFLAGS += -x musa -mtgpu --cuda-gpu-arch=mp_21 --cuda-gpu-arch=mp_22
 	else
 		ifneq ('', '$(wildcard /opt/cuda)')
 			CUDA_PATH ?= /opt/cuda

ggml/src/CMakeLists.txt

@@ -364,7 +364,7 @@ if (GGML_CUDA)
         if (GGML_MUSA)
             set_source_files_properties(${GGML_SOURCES_CUDA} PROPERTIES LANGUAGE CXX)
             foreach(SOURCE ${GGML_SOURCES_CUDA})
-                set_property(SOURCE ${SOURCE} PROPERTY COMPILE_FLAGS "-x musa -mtgpu --cuda-gpu-arch=mp_22")
+                set_property(SOURCE ${SOURCE} PROPERTY COMPILE_FLAGS "-x musa -mtgpu --cuda-gpu-arch=mp_21 --cuda-gpu-arch=mp_22")
             endforeach()
         endif()
 

ggml/src/ggml-cuda.cu

@@ -136,7 +136,7 @@ static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device)
     return res;
 #else
 
-#if !defined(GGML_USE_HIPBLAS) && !defined(GGML_USE_MUSA)
+#if !defined(GGML_USE_HIPBLAS)
     cudaError_t err;
     if (getenv("GGML_CUDA_ENABLE_UNIFIED_MEMORY") != nullptr)
     {
@@ -149,7 +149,7 @@ static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device)
     return err;
 #else
     return cudaMalloc(ptr, size);
-#endif // !defined(GGML_USE_HIPBLAS) && !defined(GGML_USE_MUSA)
+#endif // !defined(GGML_USE_HIPBLAS)
 
 #endif
 }
@@ -2351,6 +2351,7 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
             break;
         case GGML_OP_RWKV_WKV:
             ggml_cuda_op_rwkv_wkv(ctx, dst);
+            break;
         case GGML_OP_CROSS_ENTROPY_LOSS_BACK:
             ggml_cuda_cross_entropy_loss_back(ctx, dst);
             break;
@@ -2829,6 +2830,12 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
                 if (op->op == GGML_OP_MUL_MAT && a->ne[3] != b->ne[3]) {
                     return false;
                 }
+#ifdef GGML_USE_MUSA
+                if (b->type == GGML_TYPE_F16 && b->ne[2]*b->ne[3] > 1 &&
+                    !ggml_is_transposed(a) && !ggml_is_transposed(b)) {
+                    return false;
+                }
+#endif // GGML_USE_MUSA
                 switch (a->type) {
                     case GGML_TYPE_F32:
                     case GGML_TYPE_F16:
@@ -2852,6 +2859,11 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
                     case GGML_TYPE_IQ3_XXS:
                     case GGML_TYPE_IQ4_NL:
                     case GGML_TYPE_IQ4_XS:
+#ifdef GGML_USE_MUSA
+                        if (a->type == GGML_TYPE_Q3_K) {
+                            return false;
+                        }
+#endif // GGML_USE_MUSA
                         return true;
                     default:
                         return false;
@@ -2977,6 +2989,9 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
         case GGML_OP_RWKV_WKV:
             return true;
         case GGML_OP_FLASH_ATTN_EXT: {
+#ifndef FLASH_ATTN_AVAILABLE
+            return false;
+#endif
             if (op->src[0]->ne[0] ==  64 && op->src[1]->type == GGML_TYPE_F16) {
                 return true;
             }

ggml/src/ggml-cuda/common.cuh

@@ -50,6 +50,8 @@
 #define CC_RDNA1      (CC_OFFSET_AMD + 1010)
 #define CC_RDNA2      (CC_OFFSET_AMD + 1030)
 #define CC_RDNA3      (CC_OFFSET_AMD + 1100)
+#define CC_QY1        210
+#define CC_QY2        220
 
 #define MATRIX_ROW_PADDING 512 // last row of quant. matrices is a multiple of this to avoid out-of-bounds memory accesses
 
@@ -134,6 +136,10 @@ typedef float2 dfloat2;
 #define INT8_MMA_AVAILABLE
 #endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_TURING
 
+#if !(defined(GGML_USE_MUSA) && __MUSA_ARCH__ <= CC_QY1)
+#define FLASH_ATTN_AVAILABLE
+#endif // !(defined(GGML_USE_MUSA) && __MUSA_ARCH__ <= CC_QY1)
+
 static constexpr bool fast_fp16_available(const int cc) {
     return cc >= CC_PASCAL && cc != 610;
 }

ggml/src/ggml-cuda/fattn-tile-f32.cu

@@ -44,13 +44,17 @@ static __global__ void flash_attn_tile_ext_f32(
         const int ne1,
         const int ne2,
         const int ne3) {
+#ifndef FLASH_ATTN_AVAILABLE
+    NO_DEVICE_CODE;
+    return;
+#endif // FLASH_ATTN_AVAILABLE
     // Skip unused kernel variants for faster compilation:
     if (use_logit_softcap && !(D == 128 || D == 256)) {
         NO_DEVICE_CODE;
         return;
     }
 
-    //In this kernel Q, K, V are matrices while i, j, k are matrix indices.
+    // In this kernel Q, K, V are matrices while i, j, k are matrix indices.
 
     const int ic0 = (blockIdx.x / parallel_blocks) * ncols; // Index of the Q/QKV column to work on.
     const int ip  =  blockIdx.x % parallel_blocks; // Index in group of blocks running for the same column in parallel.

ggml/src/ggml-cuda/vendors/musa.h

@@ -26,6 +26,7 @@
 #define cublasSetStream mublasSetStream
 #define cublasSgemm mublasSgemm
 #define cublasStatus_t mublasStatus_t
+#define cublasOperation_t mublasOperation_t
 #define cublasGetStatusString mublasStatus_to_string
 #define cudaDataType_t musaDataType_t
 #define cudaDeviceCanAccessPeer musaDeviceCanAccessPeer
@@ -56,6 +57,7 @@
 #define cudaLaunchHostFunc musaLaunchHostFunc
 #define cudaMalloc musaMalloc
 #define cudaMallocHost musaMallocHost
+#define cudaMallocManaged musaMallocManaged
 #define cudaMemcpy musaMemcpy
 #define cudaMemcpyAsync musaMemcpyAsync
 #define cudaMemcpyPeerAsync musaMemcpyPeerAsync

tests/test-lora-conversion-inference.sh

@@ -10,11 +10,16 @@ declare -a params=(
 
 MODELS_REPO=lora-tests
 MODELS_REPO_URL=https://huggingface.co/ggml-org/$MODELS_REPO
+COMMIT=c26d5fb85b4070a9e9c4e65d132c783b98086890
 
 # Clone the Hugging Face repository if the directory does not exist
 if [ ! -d "$MODELS_REPO" ]; then
     echo "Cloning the Hugging Face repository..."
     git clone $MODELS_REPO_URL --depth 1
+    cd $MODELS_REPO
+    git fetch --depth=1 origin $COMMIT
+    git reset --hard $COMMIT
+    cd -
 else
     echo "Repository already exists. Skipping clone."
 fi