diff --git a/Makefile b/Makefile index 32598edfe..f01bf0c83 100644 --- a/Makefile +++ b/Makefile @@ -253,11 +253,6 @@ ifdef LLAMA_CUDA_KQUANTS_ITER else NVCCFLAGS += -DK_QUANTS_PER_ITERATION=2 endif -ifdef LLAMA_CUDA_MMQ_Y - NVCCFLAGS += -DGGML_CUDA_MMQ_Y=$(LLAMA_CUDA_MMQ_Y) -else - NVCCFLAGS += -DGGML_CUDA_MMQ_Y=64 -endif # LLAMA_CUDA_MMQ_Y #ifdef LLAMA_CUDA_CUBLAS # NVCCFLAGS += -DGGML_CUDA_CUBLAS #endif # LLAMA_CUDA_CUBLAS diff --git a/README.md b/README.md index 2ece294b7..6900b1152 100644 --- a/README.md +++ b/README.md @@ -406,7 +406,6 @@ Building the program with BLAS support may lead to some performance improvements ---> | Option | Legal values | Default | Description | |-------------------------|------------------------|---------|-------------| - | LLAMA_CUDA_MMQ_Y | Positive integer >= 32 | 64 | Tile size in y direction when using the custom CUDA kernels for prompt processing. Higher values can be faster depending on the amount of shared memory available. Power of 2 heavily recommended. | | LLAMA_CUDA_FORCE_DMMV | Boolean | false | Force the use of dequantization + matrix vector multiplication kernels instead of using kernels that do matrix vector multiplication on quantized data. By default the decision is made based on compute capability (MMVQ for 6.1/Pascal/GTX 1000 or higher). Does not affect k-quants. | | LLAMA_CUDA_DMMV_X | Positive integer >= 32 | 32 | Number of values in x direction processed by the CUDA dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. | | LLAMA_CUDA_MMV_Y | Positive integer | 1 | Block size in y direction for the CUDA mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants. | diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 9d42efb0d..6390b1158 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -14,6 +14,7 @@ #include "ggml.h" #define MIN_CC_DP4A 610 // minimum compute capability for __dp4a, an intrinsic for byte-wise dot products +#define CC_TURING 700 #if defined(_MSC_VER) #pragma warning(disable: 4244 4267) // possible loss of data @@ -262,10 +263,6 @@ static_assert(sizeof(block_q6_K) == sizeof(ggml_fp16_t) + 13*QK_K/16, "wrong q6_ #define CUDA_QUANTIZE_BLOCK_SIZE 256 #define CUDA_DEQUANTIZE_BLOCK_SIZE 256 -#ifndef GGML_CUDA_MMQ_Y -#define GGML_CUDA_MMQ_Y 64 -#endif // GGML_CUDA_MMQ_Y - // dmmv = dequantize_mul_mat_vec #ifndef GGML_CUDA_DMMV_X #define GGML_CUDA_DMMV_X 32 @@ -285,6 +282,20 @@ struct ggml_tensor_extra_gpu { cudaEvent_t events[GGML_CUDA_MAX_DEVICES]; // events for synchronizing multiple GPUs }; +static int g_device_count = -1; +static int g_main_device = 0; +static int g_compute_capabilities[GGML_CUDA_MAX_DEVICES]; +static float g_tensor_split[GGML_CUDA_MAX_DEVICES] = {0}; +static bool g_mul_mat_q = false; + +static void * g_scratch_buffer = nullptr; +static size_t g_scratch_size = 1024*1024*1024; // 1 GB by default +static size_t g_scratch_offset = 0; + +static cublasHandle_t g_cublas_handles[GGML_CUDA_MAX_DEVICES] = {nullptr}; + +static cudaStream_t g_cudaStreams_main[GGML_CUDA_MAX_DEVICES] = { nullptr }; + static __global__ void add_f32(const float * x, const float * y, float * dst, const int kx, const int ky) { const int i = blockDim.x*blockIdx.x + threadIdx.x; @@ -1549,8 +1560,8 @@ template static __device__ __forceinline__ float vec_dot_q8_1_q8_1_imp #else const float2 dm8f = __half22float2(dm8); const float2 ds8f = __half22float2(ds8); - const float d8d8 = dm8.x * ds8.x; - const float m8s8 = dm8.y * ds8.y; + const float d8d8 = dm8f.x * ds8f.x; + const float m8s8 = dm8f.y * ds8f.y; #endif // GGML_CUDA_F16 // scale second part of sum by QI8_1/ vdr to compensate for multiple threads adding it @@ -1884,21 +1895,21 @@ static __device__ __forceinline__ float vec_dot_q4_0_q8_1( return vec_dot_q4_0_q8_1_impl(v, u, bq4_0->d, bq8_1->ds); } -static __device__ __forceinline__ void allocate_tiles_q4_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q4_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_qs[GGML_CUDA_MMQ_Y * (WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ float tile_x_d[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI4_0) + GGML_CUDA_MMQ_Y/QI4_0]; + __shared__ int tile_x_qs[mmq_y * (WARP_SIZE) + mmq_y]; + __shared__ float tile_x_d[mmq_y * (WARP_SIZE/QI4_0) + mmq_y/QI4_0]; *x_ql = tile_x_qs; *x_dm = (half2 *) tile_x_d; } -template static __device__ __forceinline__ void load_tiles_q4_0( +template static __device__ __forceinline__ void load_tiles_q4_0( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -1910,7 +1921,7 @@ template static __device__ __forceinline__ void load_tiles_q4_ float * x_dmf = (float *) x_dm; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -1920,39 +1931,30 @@ template static __device__ __forceinline__ void load_tiles_q4_ const block_q4_0 * bxi = bx0 + i*blocks_per_row + kbx; x_ql[i * (WARP_SIZE + 1) + k] = get_int_from_uint8(bxi->qs, kqsx); - x_dmf[i * (WARP_SIZE/QI4_0) + i / QI4_0 + kbx] = bxi->d; + // x_dmf[i * (WARP_SIZE/QI4_0) + i / QI4_0 + kbx] = bxi->d; } -// const int blocks_per_tile_x_row = WARP_SIZE / QI4_0; -// const int kbxd = k % blocks_per_tile_x_row; + const int blocks_per_tile_x_row = WARP_SIZE / QI4_0; + const int kbxd = k % blocks_per_tile_x_row; -// #pragma unroll -// for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI4_0) { -// FIXME out-of-bounds -// const int i = i0 + i_offset * QI4_0 + k / blocks_per_tile_x_row; +#pragma unroll + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI4_0) { + int i = i0 + i_offset * QI4_0 + k / blocks_per_tile_x_row; -// if (i >= GGML_CUDA_MMQ_Y) { -// return; -// } + if (need_check) { + i = min(i, i_max); + } -// const block_q4_0 * bxi = bx0 + i*blocks_per_row + kbxd; + const block_q4_0 * bxi = bx0 + i*blocks_per_row + kbxd; -// x_dm[i * (WARP_SIZE/QI4_0) + i / QI4_0 + kbxd].x = bxi->d; -// } + x_dmf[i * (WARP_SIZE/QI4_0) + i / QI4_0 + kbxd] = bxi->d; + } } static __device__ __forceinline__ float vec_dot_q4_0_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q4_0_Q8_1_MMQ == 0); - const int kyqs = k % (QI8_1/2) + QI8_1 * (k / (QI8_1/2)); const float * x_dmf = (float *) x_dm; @@ -1960,13 +1962,13 @@ static __device__ __forceinline__ float vec_dot_q4_0_q8_1_mul_mat( #pragma unroll for (int l = 0; l < VDR_Q4_0_Q8_1_MMQ; ++l) { - u[2*l+0] = y_qs[j * (2*WARP_SIZE) + kyqs + l]; - u[2*l+1] = y_qs[j * (2*WARP_SIZE) + kyqs + l + QI4_0]; + u[2*l+0] = y_qs[j * WARP_SIZE + (kyqs + l) % WARP_SIZE]; + u[2*l+1] = y_qs[j * WARP_SIZE + (kyqs + l + QI4_0) % WARP_SIZE]; } return vec_dot_q4_0_q8_1_impl (&x_ql[i * (WARP_SIZE + 1) + k], u, x_dmf[i * (WARP_SIZE/QI4_0) + i/QI4_0 + k/QI4_0], - y_ds[j * (2*WARP_SIZE/QI8_1) + 2*k/QI8_1]); + y_ds[j * (WARP_SIZE/QI8_1) + (2*k/QI8_1) % (WARP_SIZE/QI8_1)]); } static __device__ __forceinline__ float vec_dot_q4_1_q8_1( @@ -1987,21 +1989,21 @@ static __device__ __forceinline__ float vec_dot_q4_1_q8_1( return vec_dot_q4_1_q8_1_impl(v, u, bq4_1->dm, bq8_1->ds); } -static __device__ __forceinline__ void allocate_tiles_q4_1(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q4_1(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_qs[GGML_CUDA_MMQ_Y * (WARP_SIZE) + + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI4_1) + GGML_CUDA_MMQ_Y/QI4_1]; + __shared__ int tile_x_qs[mmq_y * (WARP_SIZE) + + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI4_1) + mmq_y/QI4_1]; *x_ql = tile_x_qs; *x_dm = tile_x_dm; } -template static __device__ __forceinline__ void load_tiles_q4_1( +template static __device__ __forceinline__ void load_tiles_q4_1( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2011,7 +2013,7 @@ template static __device__ __forceinline__ void load_tiles_q4_ const block_q4_1 * bx0 = (block_q4_1 *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2027,7 +2029,7 @@ template static __device__ __forceinline__ void load_tiles_q4_ const int kbxd = k % blocks_per_tile_x_row; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI4_1) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI4_1) { int i = i0 + i_offset * QI4_1 + k / blocks_per_tile_x_row; if (need_check) { @@ -2044,27 +2046,19 @@ static __device__ __forceinline__ float vec_dot_q4_1_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q4_1_Q8_1_MMQ == 0); - const int kyqs = k % (QI8_1/2) + QI8_1 * (k / (QI8_1/2)); int u[2*VDR_Q4_1_Q8_1_MMQ]; #pragma unroll for (int l = 0; l < VDR_Q4_1_Q8_1_MMQ; ++l) { - u[2*l+0] = y_qs[j * (2*WARP_SIZE) + kyqs + l]; - u[2*l+1] = y_qs[j * (2*WARP_SIZE) + kyqs + l + QI4_1]; + u[2*l+0] = y_qs[j * WARP_SIZE + (kyqs + l) % WARP_SIZE]; + u[2*l+1] = y_qs[j * WARP_SIZE + (kyqs + l + QI4_1) % WARP_SIZE]; } return vec_dot_q4_1_q8_1_impl (&x_ql[i * (WARP_SIZE + 1) + k], u, x_dm[i * (WARP_SIZE/QI4_1) + i/QI4_1 + k/QI4_1], - y_ds[j * (2*WARP_SIZE/QI8_1) + 2*k/QI8_1]); + y_ds[j * (WARP_SIZE/QI8_1) + (2*k/QI8_1) % (WARP_SIZE/QI8_1)]); } static __device__ __forceinline__ float vec_dot_q5_0_q8_1( @@ -2087,21 +2081,21 @@ static __device__ __forceinline__ float vec_dot_q5_0_q8_1( return vec_dot_q5_0_q8_1_impl(vl, vh, u, bq5_0->d, bq8_1->ds); } -static __device__ __forceinline__ void allocate_tiles_q5_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q5_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (2*WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ float tile_x_d[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI5_0) + GGML_CUDA_MMQ_Y/QI5_0]; + __shared__ int tile_x_ql[mmq_y * (2*WARP_SIZE) + mmq_y]; + __shared__ float tile_x_d[mmq_y * (WARP_SIZE/QI5_0) + mmq_y/QI5_0]; *x_ql = tile_x_ql; *x_dm = (half2 *) tile_x_d; } -template static __device__ __forceinline__ void load_tiles_q5_0( +template static __device__ __forceinline__ void load_tiles_q5_0( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2111,7 +2105,7 @@ template static __device__ __forceinline__ void load_tiles_q5_ const block_q5_0 * bx0 = (block_q5_0 *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2147,7 +2141,7 @@ template static __device__ __forceinline__ void load_tiles_q5_ float * x_dmf = (float *) x_dm; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI5_0) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI5_0) { int i = i0 + i_offset * QI5_0 + k / blocks_per_tile_x_row; if (need_check) { @@ -2164,14 +2158,6 @@ static __device__ __forceinline__ float vec_dot_q5_0_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q5_0_Q8_1_MMQ == 0); - const int kyqs = k % (QI8_1/2) + QI8_1 * (k / (QI8_1/2)); const int index_bx = i * (WARP_SIZE/QI5_0) + i/QI5_0 + k/QI5_0; const float * x_dmf = (const float *) x_dm; @@ -2181,12 +2167,12 @@ static __device__ __forceinline__ float vec_dot_q5_0_q8_1_mul_mat( #pragma unroll for (int l = 0; l < VDR_Q5_0_Q8_1_MMQ; ++l) { - u[2*l+0] = y_qs[j * (2*WARP_SIZE) + kyqs + l]; - u[2*l+1] = y_qs[j * (2*WARP_SIZE) + kyqs + l + QI5_0]; + u[2*l+0] = y_qs[j * WARP_SIZE + (kyqs + l) % WARP_SIZE]; + u[2*l+1] = y_qs[j * WARP_SIZE + (kyqs + l + QI5_0) % WARP_SIZE]; } return vec_dot_q8_0_q8_1_impl - (&x_ql[i * (2*WARP_SIZE + 1) + 2 * k], u, x_dmf[index_bx], y_df[j * (2*WARP_SIZE/QI8_1) + 2*k/QI8_1]); + (&x_ql[i * (2*WARP_SIZE + 1) + 2 * k], u, x_dmf[index_bx], y_df[j * (WARP_SIZE/QI8_1) + (2*k/QI8_1) % (WARP_SIZE/QI8_1)]); } static __device__ __forceinline__ float vec_dot_q5_1_q8_1( @@ -2209,21 +2195,21 @@ static __device__ __forceinline__ float vec_dot_q5_1_q8_1( return vec_dot_q5_1_q8_1_impl(vl, vh, u, bq5_1->dm, bq8_1->ds); } -static __device__ __forceinline__ void allocate_tiles_q5_1(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q5_1(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (2*WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI5_1) + GGML_CUDA_MMQ_Y/QI5_1]; + __shared__ int tile_x_ql[mmq_y * (2*WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI5_1) + mmq_y/QI5_1]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; } -template static __device__ __forceinline__ void load_tiles_q5_1( +template static __device__ __forceinline__ void load_tiles_q5_1( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2233,7 +2219,7 @@ template static __device__ __forceinline__ void load_tiles_q5_ const block_q5_1 * bx0 = (block_q5_1 *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2266,7 +2252,7 @@ template static __device__ __forceinline__ void load_tiles_q5_ const int kbxd = k % blocks_per_tile_x_row; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI5_1) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI5_1) { int i = i0 + i_offset * QI5_1 + k / blocks_per_tile_x_row; if (need_check) { @@ -2283,14 +2269,6 @@ static __device__ __forceinline__ float vec_dot_q5_1_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q5_1_Q8_1_MMQ == 0); - const int kyqs = k % (QI8_1/2) + QI8_1 * (k / (QI8_1/2)); const int index_bx = i * (WARP_SIZE/QI5_1) + + i/QI5_1 + k/QI5_1; @@ -2298,12 +2276,12 @@ static __device__ __forceinline__ float vec_dot_q5_1_q8_1_mul_mat( #pragma unroll for (int l = 0; l < VDR_Q5_1_Q8_1_MMQ; ++l) { - u[2*l+0] = y_qs[j * (2*WARP_SIZE) + kyqs + l]; - u[2*l+1] = y_qs[j * (2*WARP_SIZE) + kyqs + l + QI5_1]; + u[2*l+0] = y_qs[j * WARP_SIZE + (kyqs + l) % WARP_SIZE]; + u[2*l+1] = y_qs[j * WARP_SIZE + (kyqs + l + QI5_1) % WARP_SIZE]; } return vec_dot_q8_1_q8_1_impl - (&x_ql[i * (2*WARP_SIZE + 1) + 2 * k], u, x_dm[index_bx], y_ds[j * (2*WARP_SIZE/QI8_1) + 2*k/QI8_1]); + (&x_ql[i * (2*WARP_SIZE + 1) + 2 * k], u, x_dm[index_bx], y_ds[j * (WARP_SIZE/QI8_1) + (2*k/QI8_1) % (WARP_SIZE/QI8_1)]); } static __device__ __forceinline__ float vec_dot_q8_0_q8_1( @@ -2323,21 +2301,21 @@ static __device__ __forceinline__ float vec_dot_q8_0_q8_1( return vec_dot_q8_0_q8_1_impl(v, u, bq8_0->d, bq8_1->ds.x); } -static __device__ __forceinline__ void allocate_tiles_q8_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q8_0(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_qs[GGML_CUDA_MMQ_Y * (WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ float tile_x_d[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI8_0) + GGML_CUDA_MMQ_Y/QI8_0]; + __shared__ int tile_x_qs[mmq_y * (WARP_SIZE) + mmq_y]; + __shared__ float tile_x_d[mmq_y * (WARP_SIZE/QI8_0) + mmq_y/QI8_0]; *x_ql = tile_x_qs; *x_dm = (half2 *) tile_x_d; } -template static __device__ __forceinline__ void load_tiles_q8_0( +template static __device__ __forceinline__ void load_tiles_q8_0( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2348,7 +2326,7 @@ template static __device__ __forceinline__ void load_tiles_q8_ const block_q8_0 * bx0 = (block_q8_0 *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2358,41 +2336,29 @@ template static __device__ __forceinline__ void load_tiles_q8_ const block_q8_0 * bxi = bx0 + i*blocks_per_row + kbx; x_ql[i * (WARP_SIZE + 1) + k] = get_int_from_int8(bxi->qs, kqsx); - x_dmf[i * (WARP_SIZE/QI8_0) + i / QI8_0 + kbx] = bxi->d; } -// const int blocks_per_tile_x_row = WARP_SIZE / QI8_0; -// const int kbxd = k % blocks_per_tile_x_row; + const int blocks_per_tile_x_row = WARP_SIZE / QI8_0; + const int kbxd = k % blocks_per_tile_x_row; -// #pragma unroll -// for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI8_0) { -// FIXME out-of-bounds -// const int i = i0 + i_offset * QI8_0 + k / blocks_per_tile_x_row; +#pragma unroll + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI8_0) { + int i = i0 + i_offset * QI8_0 + k / blocks_per_tile_x_row; -// #if GGML_CUDA_MMQ_Y < 64 -// if (i >= GGML_CUDA_MMQ_Y) { -// return; -// } -// #endif // GGML_CUDA_MMQ_Y < 64 + if (need_check) { + i = min(i, i_max); + } -// const block_q8_0 * bxi = bx0 + i*blocks_per_row + kbxd; + const block_q8_0 * bxi = bx0 + i*blocks_per_row + kbxd; -// x_dm[i * (WARP_SIZE/QI8_0) + i / QI8_0 + kbxd].x = bxi->d; -// } + x_dmf[i * (WARP_SIZE/QI8_0) + i / QI8_0 + kbxd] = bxi->d; + } } static __device__ __forceinline__ float vec_dot_q8_0_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q8_0_Q8_1_MMQ == 0); - const float * x_dmf = (const float *) x_dm; const float * y_df = (const float *) y_ds; @@ -2424,23 +2390,23 @@ static __device__ __forceinline__ float vec_dot_q2_K_q8_1( return vec_dot_q2_K_q8_1_impl_mmvq(v, u, scales, bq2_K->dm, d8); } -static __device__ __forceinline__ void allocate_tiles_q2_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q2_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI2_K) + GGML_CUDA_MMQ_Y/QI2_K]; - __shared__ int tile_x_sc[GGML_CUDA_MMQ_Y * (WARP_SIZE/4) + GGML_CUDA_MMQ_Y/4]; + __shared__ int tile_x_ql[mmq_y * (WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI2_K) + mmq_y/QI2_K]; + __shared__ int tile_x_sc[mmq_y * (WARP_SIZE/4) + mmq_y/4]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; *x_sc = tile_x_sc; } -template static __device__ __forceinline__ void load_tiles_q2_K( +template static __device__ __forceinline__ void load_tiles_q2_K( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2450,7 +2416,7 @@ template static __device__ __forceinline__ void load_tiles_q2_ const block_q2_K * bx0 = (block_q2_K *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2466,8 +2432,8 @@ template static __device__ __forceinline__ void load_tiles_q2_ const int kbxd = k % blocks_per_tile_x_row; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI2_K) { - int i = (i0 + i_offset * QI2_K + k / blocks_per_tile_x_row) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI2_K) { + int i = (i0 + i_offset * QI2_K + k / blocks_per_tile_x_row) % mmq_y; if (need_check) { i = min(i, i_max); @@ -2479,7 +2445,7 @@ template static __device__ __forceinline__ void load_tiles_q2_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 4) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 4) { int i = i0 + i_offset * 4 + k / (WARP_SIZE/4); if (need_check) { @@ -2496,14 +2462,6 @@ static __device__ __forceinline__ float vec_dot_q2_K_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q2_K_Q8_1_MMQ == 0); - const int kbx = k / QI2_K; const int ky = (k % QI2_K) * QR2_K; const float * y_df = (const float *) y_ds; @@ -2520,7 +2478,7 @@ static __device__ __forceinline__ float vec_dot_q2_K_q8_1_mul_mat( const uint8_t * scales = ((const uint8_t *) &x_sc[i * (WARP_SIZE/4) + i/4 + kbx*4]) + ky/4; - const int index_y = j * (QR2_K*WARP_SIZE) + QR2_K*k; + const int index_y = j * WARP_SIZE + (QR2_K*k) % WARP_SIZE; return vec_dot_q2_K_q8_1_impl_mmq(v, &y_qs[index_y], scales, x_dm[i * (WARP_SIZE/QI2_K) + i/QI2_K + kbx], y_df[index_y/QI8_1]); } @@ -2551,12 +2509,12 @@ static __device__ __forceinline__ float vec_dot_q3_K_q8_1( return vec_dot_q3_K_q8_1_impl_mmvq(vl, vh, u, bq3_K->scales, scale_offset, d, d8); } -static __device__ __forceinline__ void allocate_tiles_q3_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q3_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI3_K) + GGML_CUDA_MMQ_Y/QI3_K]; - __shared__ int tile_x_qh[GGML_CUDA_MMQ_Y * (WARP_SIZE/2) + GGML_CUDA_MMQ_Y/2]; - __shared__ int tile_x_sc[GGML_CUDA_MMQ_Y * (WARP_SIZE/4) + GGML_CUDA_MMQ_Y/4]; + __shared__ int tile_x_ql[mmq_y * (WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI3_K) + mmq_y/QI3_K]; + __shared__ int tile_x_qh[mmq_y * (WARP_SIZE/2) + mmq_y/2]; + __shared__ int tile_x_sc[mmq_y * (WARP_SIZE/4) + mmq_y/4]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; @@ -2564,12 +2522,12 @@ static __device__ __forceinline__ void allocate_tiles_q3_K(int ** x_ql, half2 ** *x_sc = tile_x_sc; } -template static __device__ __forceinline__ void load_tiles_q3_K( +template static __device__ __forceinline__ void load_tiles_q3_K( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2579,7 +2537,7 @@ template static __device__ __forceinline__ void load_tiles_q3_ const block_q3_K * bx0 = (block_q3_K *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2596,8 +2554,8 @@ template static __device__ __forceinline__ void load_tiles_q3_ float * x_dmf = (float *) x_dm; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI3_K) { - int i = (i0 + i_offset * QI3_K + k / blocks_per_tile_x_row) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI3_K) { + int i = (i0 + i_offset * QI3_K + k / blocks_per_tile_x_row) % mmq_y; if (need_check) { i = min(i, i_max); @@ -2609,7 +2567,7 @@ template static __device__ __forceinline__ void load_tiles_q3_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 2) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 2) { int i = i0 + i_offset * 2 + k / (WARP_SIZE/2); if (need_check) { @@ -2623,7 +2581,7 @@ template static __device__ __forceinline__ void load_tiles_q3_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 4) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 4) { int i = i0 + i_offset * 4 + k / (WARP_SIZE/4); if (need_check) { @@ -2652,14 +2610,6 @@ static __device__ __forceinline__ float vec_dot_q3_K_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q3_K_Q8_1_MMQ == 0); - const int kbx = k / QI3_K; const int ky = (k % QI3_K) * QR3_K; const float * x_dmf = (const float *) x_dm; @@ -2681,7 +2631,7 @@ static __device__ __forceinline__ float vec_dot_q3_K_q8_1_mul_mat( v[l] = __vsubss4(vll, vlh); } - const int index_y = j * (QR3_K*WARP_SIZE) + k*QR3_K; + const int index_y = j * WARP_SIZE + (k*QR3_K) % WARP_SIZE; return vec_dot_q3_K_q8_1_impl_mmq(v, &y_qs[index_y], scales, x_dmf[i * (WARP_SIZE/QI3_K) + i/QI3_K + kbx], y_df[index_y/QI8_1]); } @@ -2778,23 +2728,23 @@ static __device__ __forceinline__ float vec_dot_q4_K_q8_1( #endif } -static __device__ __forceinline__ void allocate_tiles_q4_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q4_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI4_K) + GGML_CUDA_MMQ_Y/QI4_K]; - __shared__ int tile_x_sc[GGML_CUDA_MMQ_Y * (WARP_SIZE/8) + GGML_CUDA_MMQ_Y/8]; + __shared__ int tile_x_ql[mmq_y * (WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI4_K) + mmq_y/QI4_K]; + __shared__ int tile_x_sc[mmq_y * (WARP_SIZE/8) + mmq_y/8]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; *x_sc = tile_x_sc; } -template static __device__ __forceinline__ void load_tiles_q4_K( +template static __device__ __forceinline__ void load_tiles_q4_K( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2804,7 +2754,7 @@ template static __device__ __forceinline__ void load_tiles_q4_ const block_q4_K * bx0 = (block_q4_K *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -2820,8 +2770,8 @@ template static __device__ __forceinline__ void load_tiles_q4_ const int kbxd = k % blocks_per_tile_x_row; // == 0 if QK_K == 256 #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI4_K) { - int i = (i0 + i_offset * QI4_K + k / blocks_per_tile_x_row) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI4_K) { + int i = (i0 + i_offset * QI4_K + k / blocks_per_tile_x_row) % mmq_y; if (need_check) { i = min(i, i_max); @@ -2833,8 +2783,8 @@ template static __device__ __forceinline__ void load_tiles_q4_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 8) { - int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 8) { + int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % mmq_y; if (need_check) { i = min(i, i_max); @@ -2858,14 +2808,6 @@ static __device__ __forceinline__ float vec_dot_q4_K_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q4_K_Q8_1_MMQ == 0); - int v[QR4_K*VDR_Q4_K_Q8_1_MMQ]; #pragma unroll @@ -2876,7 +2818,7 @@ static __device__ __forceinline__ float vec_dot_q4_K_q8_1_mul_mat( const uint8_t * sc = ((const uint8_t *) &x_sc[i * (WARP_SIZE/8) + i/8 + k/16]) + 2*((k % 16) / 8); - const int index_y = j * (QR4_K*WARP_SIZE) + QR4_K*k; + const int index_y = j * WARP_SIZE + (QR4_K*k) % WARP_SIZE; return vec_dot_q4_K_q8_1_impl_mmq(v, &y_qs[index_y], sc, sc+8, x_dm[i * (WARP_SIZE/QI4_K) + i/QI4_K], &y_ds[index_y/QI8_1]); } @@ -2969,23 +2911,23 @@ static __device__ __forceinline__ float vec_dot_q5_K_q8_1( #endif } -static __device__ __forceinline__ void allocate_tiles_q5_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q5_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (2*WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI5_K) + GGML_CUDA_MMQ_Y/QI5_K]; - __shared__ int tile_x_sc[GGML_CUDA_MMQ_Y * (WARP_SIZE/8) + GGML_CUDA_MMQ_Y/8]; + __shared__ int tile_x_ql[mmq_y * (2*WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI5_K) + mmq_y/QI5_K]; + __shared__ int tile_x_sc[mmq_y * (WARP_SIZE/8) + mmq_y/8]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; *x_sc = tile_x_sc; } -template static __device__ __forceinline__ void load_tiles_q5_K( +template static __device__ __forceinline__ void load_tiles_q5_K( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -2995,7 +2937,7 @@ template static __device__ __forceinline__ void load_tiles_q5_ const block_q5_K * bx0 = (block_q5_K *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -3024,8 +2966,8 @@ template static __device__ __forceinline__ void load_tiles_q5_ const int kbxd = k % blocks_per_tile_x_row; // == 0 if QK_K == 256 #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI5_K) { - int i = (i0 + i_offset * QI5_K + k / blocks_per_tile_x_row) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI5_K) { + int i = (i0 + i_offset * QI5_K + k / blocks_per_tile_x_row) % mmq_y; if (need_check) { i = min(i, i_max); @@ -3037,8 +2979,8 @@ template static __device__ __forceinline__ void load_tiles_q5_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 8) { - int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 8) { + int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % mmq_y; if (need_check) { i = min(i, i_max); @@ -3062,18 +3004,10 @@ static __device__ __forceinline__ float vec_dot_q5_K_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q5_K_Q8_1_MMQ == 0); - const uint8_t * sc = ((const uint8_t *) &x_sc[i * (WARP_SIZE/8) + i/8 + k/16]) + 2 * ((k % 16) / 8); - const int index_x = i * (QR5_K*WARP_SIZE + 1) + QR5_K*k; - const int index_y = j * (QR5_K*WARP_SIZE) + QR5_K*k; + const int index_x = i * (QR5_K*WARP_SIZE + 1) + QR5_K*k; + const int index_y = j * WARP_SIZE + (QR5_K*k) % WARP_SIZE; return vec_dot_q4_K_q8_1_impl_mmq(&x_ql[index_x], &y_qs[index_y], sc, sc+8, x_dm[i * (WARP_SIZE/QI5_K) + i/QI5_K], &y_ds[index_y/QI8_1]); } @@ -3103,23 +3037,23 @@ static __device__ __forceinline__ float vec_dot_q6_K_q8_1( return vec_dot_q6_K_q8_1_impl_mmvq(vl, vh, u, scales, bq6_K->d, d8); } -static __device__ __forceinline__ void allocate_tiles_q6_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { +template static __device__ __forceinline__ void allocate_tiles_q6_K(int ** x_ql, half2 ** x_dm, int ** x_qh, int ** x_sc) { - __shared__ int tile_x_ql[GGML_CUDA_MMQ_Y * (2*WARP_SIZE) + GGML_CUDA_MMQ_Y]; - __shared__ half2 tile_x_dm[GGML_CUDA_MMQ_Y * (WARP_SIZE/QI6_K) + GGML_CUDA_MMQ_Y/QI6_K]; - __shared__ int tile_x_sc[GGML_CUDA_MMQ_Y * (WARP_SIZE/8) + GGML_CUDA_MMQ_Y/8]; + __shared__ int tile_x_ql[mmq_y * (2*WARP_SIZE) + mmq_y]; + __shared__ half2 tile_x_dm[mmq_y * (WARP_SIZE/QI6_K) + mmq_y/QI6_K]; + __shared__ int tile_x_sc[mmq_y * (WARP_SIZE/8) + mmq_y/8]; *x_ql = tile_x_ql; *x_dm = tile_x_dm; *x_sc = tile_x_sc; } -template static __device__ __forceinline__ void load_tiles_q6_K( +template static __device__ __forceinline__ void load_tiles_q6_K( const void * __restrict__ vx, int * __restrict__ x_ql, half2 * __restrict__ x_dm, int * __restrict__ x_qh, int * __restrict__ x_sc, const int & i_offset, const int & i_max, const int & k, const int & blocks_per_row) { __builtin_assume(i_offset >= 0); - __builtin_assume(i_offset < 8); + __builtin_assume(i_offset < nwarps); __builtin_assume(k >= 0); __builtin_assume(k < WARP_SIZE); @@ -3129,7 +3063,7 @@ template static __device__ __forceinline__ void load_tiles_q6_ const block_q6_K * bx0 = (block_q6_K *) vx; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8) { + for (int i0 = 0; i0 < mmq_y; i0 += nwarps) { int i = i0 + i_offset; if (need_check) { @@ -3159,8 +3093,8 @@ template static __device__ __forceinline__ void load_tiles_q6_ float * x_dmf = (float *) x_dm; #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * QI6_K) { - int i = (i0 + i_offset * QI6_K + k / blocks_per_tile_x_row) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * QI6_K) { + int i = (i0 + i_offset * QI6_K + k / blocks_per_tile_x_row) % mmq_y; if (need_check) { i = min(i, i_max); @@ -3172,8 +3106,8 @@ template static __device__ __forceinline__ void load_tiles_q6_ } #pragma unroll - for (int i0 = 0; i0 < GGML_CUDA_MMQ_Y; i0 += 8 * 8) { - int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % GGML_CUDA_MMQ_Y; + for (int i0 = 0; i0 < mmq_y; i0 += nwarps * 8) { + int i = (i0 + i_offset * 8 + k / (WARP_SIZE/8)) % mmq_y; if (need_check) { i = min(i, i_max); @@ -3189,25 +3123,17 @@ static __device__ __forceinline__ float vec_dot_q6_K_q8_1_mul_mat( const int * __restrict__ x_ql, const half2 * __restrict__ x_dm, const int * __restrict__ x_qh, const int * __restrict__ x_sc, const int * __restrict__ y_qs, const half2 * __restrict__ y_ds, const int & i, const int & j, const int & k) { - __builtin_assume(i >= 0); - __builtin_assume(i < GGML_CUDA_MMQ_Y); - __builtin_assume(j >= 0); - __builtin_assume(j < WARP_SIZE); - __builtin_assume(k >= 0); - __builtin_assume(k < WARP_SIZE); - __builtin_assume(k % VDR_Q6_K_Q8_1_MMQ == 0); - const float * x_dmf = (const float *) x_dm; const float * y_df = (const float *) y_ds; const int8_t * sc = ((const int8_t *) &x_sc[i * (WARP_SIZE/8) + i/8 + k/8]); - const int index_x = i * (QR6_K*WARP_SIZE + 1) + QR6_K*k; - const int index_y = j * (QR6_K*WARP_SIZE) + QR6_K*k; + const int index_x = i * (QR6_K*WARP_SIZE + 1) + QR6_K*k; + const int index_y = j * WARP_SIZE + (QR6_K*k) % WARP_SIZE; return vec_dot_q6_K_q8_1_impl_mmq(&x_ql[index_x], &y_qs[index_y], sc, x_dmf[i * (WARP_SIZE/QI6_K) + i/QI6_K], &y_df[index_y/QI8_1]); } -template static __global__ void mul_mat_q( const void * __restrict__ vx, const void * __restrict__ vy, float * __restrict__ dst, @@ -3222,14 +3148,11 @@ static __global__ void mul_mat_q( const int & ncols_dst = ncols_y; - const int tid_x = threadIdx.x; - const int tid_y = threadIdx.y; - - const int row_dst_0 = blockIdx.x*GGML_CUDA_MMQ_Y; + const int row_dst_0 = blockIdx.x*mmq_y; const int & row_x_0 = row_dst_0; - const int row_dst = row_dst_0 + tid_x; + const int row_dst = row_dst_0 + threadIdx.x; - const int col_dst_0 = blockIdx.y*WARP_SIZE; + const int col_dst_0 = blockIdx.y*mmq_x; const int & col_y_0 = col_dst_0; int * tile_x_ql = nullptr; @@ -3239,64 +3162,65 @@ static __global__ void mul_mat_q( allocate_tiles(&tile_x_ql, &tile_x_dm, &tile_x_qh, &tile_x_sc); - const int blocks_per_tile_y_col = qr*WARP_SIZE/QI8_1; + __shared__ int tile_y_qs[mmq_x * WARP_SIZE]; + __shared__ half2 tile_y_ds[mmq_x * WARP_SIZE/QI8_1]; - __shared__ int tile_y_qs[(WARP_SIZE) * (qr*WARP_SIZE)]; - __shared__ half2 tile_y_ds[(WARP_SIZE) * blocks_per_tile_y_col]; - - float sum[GGML_CUDA_MMQ_Y/WARP_SIZE][4] = {0.0f}; + float sum[mmq_y/WARP_SIZE][mmq_x/nwarps] = {0.0f}; for (int ib0 = 0; ib0 < blocks_per_row_x; ib0 += blocks_per_warp) { load_tiles(x + row_x_0*blocks_per_row_x + ib0, tile_x_ql, tile_x_dm, tile_x_qh, tile_x_sc, - tid_y, nrows_x-row_x_0-1, tid_x, blocks_per_row_x); + threadIdx.y, nrows_x-row_x_0-1, threadIdx.x, blocks_per_row_x); +#pragma unroll for (int ir = 0; ir < qr; ++ir) { - const int kqs = ir*WARP_SIZE + tid_x; + const int kqs = ir*WARP_SIZE + threadIdx.x; const int kbxd = kqs / QI8_1; - for (int i = 0; i < WARP_SIZE; i += 8) { - const int col_y_eff = min(col_y_0 + tid_y + i, ncols_y-1); // to prevent out-of-bounds memory accesses +#pragma unroll + for (int i = 0; i < mmq_x; i += nwarps) { + const int col_y_eff = min(col_y_0 + threadIdx.y + i, ncols_y-1); // to prevent out-of-bounds memory accesses const block_q8_1 * by0 = &y[col_y_eff*blocks_per_col_y + ib0 * (qk/QK8_1) + kbxd]; - tile_y_qs[(tid_y + i) * (qr*WARP_SIZE) + kqs] = get_int_from_int8_aligned(by0->qs, tid_x % QI8_1); + const int index_y = (threadIdx.y + i) * WARP_SIZE + kqs % WARP_SIZE; + tile_y_qs[index_y] = get_int_from_int8_aligned(by0->qs, threadIdx.x % QI8_1); } - } - for (int ids0 = 0; ids0 < WARP_SIZE; ids0 += 8 * (WARP_SIZE/blocks_per_tile_y_col)) { - const int ids = (ids0 + tid_y * (WARP_SIZE/blocks_per_tile_y_col) + tid_x / blocks_per_tile_y_col) % WARP_SIZE; - const int kby = tid_x % blocks_per_tile_y_col; - const int col_y_eff = min(col_y_0 + ids, ncols_y-1); - - // if the sum is not needed it's faster to transform the scale to f32 ahead of time - const half2 * dsi_src = &y[col_y_eff*blocks_per_col_y + ib0 * (qk/QK8_1) + kby].ds; - half2 * dsi_dst = &tile_y_ds[ids * (qr*WARP_SIZE/QI8_1) + kby]; - if (need_sum) { - *dsi_dst = *dsi_src; - } else { - float * dfi_dst = (float *) dsi_dst; - *dfi_dst = (*dsi_src).x; - } - } - - __syncthreads(); - -#if __CUDA_ARCH__ >= 700 // Unrolling the loop is slower on Pascal #pragma unroll -#endif // __CUDA_ARCH__ >= 700 - for (int k = 0; k < WARP_SIZE; k += vdr) { -#pragma unroll - for (int j = 0; j < WARP_SIZE; j += 8) { -#pragma unroll - for (int i = 0; i < GGML_CUDA_MMQ_Y; i += WARP_SIZE) { - sum[i/WARP_SIZE][j/8] += vec_dot(tile_x_ql, tile_x_dm, tile_x_qh, tile_x_sc, tile_y_qs, tile_y_ds, - tid_x + i, tid_y + j, k); + for (int ids0 = 0; ids0 < mmq_x; ids0 += nwarps * QI8_1) { + const int ids = (ids0 + threadIdx.y * QI8_1 + threadIdx.x / (WARP_SIZE/QI8_1)) % mmq_x; + const int kby = threadIdx.x % (WARP_SIZE/QI8_1); + const int col_y_eff = min(col_y_0 + ids, ncols_y-1); + + // if the sum is not needed it's faster to transform the scale to f32 ahead of time + const half2 * dsi_src = &y[col_y_eff*blocks_per_col_y + ib0 * (qk/QK8_1) + ir*(WARP_SIZE/QI8_1) + kby].ds; + half2 * dsi_dst = &tile_y_ds[ids * (WARP_SIZE/QI8_1) + kby]; + if (need_sum) { + *dsi_dst = *dsi_src; + } else { + float * dfi_dst = (float *) dsi_dst; + *dfi_dst = (*dsi_src).x; } } - } - __syncthreads(); + __syncthreads(); + +// #pragma unroll // unrolling this loop causes too much register pressure + for (int k = ir*WARP_SIZE/qr; k < (ir+1)*WARP_SIZE/qr; k += vdr) { +#pragma unroll + for (int j = 0; j < mmq_x; j += nwarps) { +#pragma unroll + for (int i = 0; i < mmq_y; i += WARP_SIZE) { + sum[i/WARP_SIZE][j/nwarps] += vec_dot( + tile_x_ql, tile_x_dm, tile_x_qh, tile_x_sc, tile_y_qs, tile_y_ds, + threadIdx.x + i, threadIdx.y + j, k); + } + } + } + + __syncthreads(); + } } @@ -3304,15 +3228,15 @@ static __global__ void mul_mat_q( return; } - for (int j = 0; j < WARP_SIZE; j += 8) { - const int col_dst = col_dst_0 + j + tid_y; + for (int j = 0; j < mmq_x; j += nwarps) { + const int col_dst = col_dst_0 + j + threadIdx.y; if (col_dst >= ncols_dst) { return; } - for (int i = 0; i < GGML_CUDA_MMQ_Y; i += WARP_SIZE) { - dst[col_dst*nrows_dst + row_dst + i] = sum[i/WARP_SIZE][j/8]; + for (int i = 0; i < mmq_y; i += WARP_SIZE) { + dst[col_dst*nrows_dst + row_dst + i] = sum[i/WARP_SIZE][j/nwarps]; } } } @@ -4014,17 +3938,52 @@ static void ggml_mul_mat_q4_0_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_0, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_0, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_0, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_0, VDR_Q4_0_Q8_1_MMQ, vec_dot_q4_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4032,17 +3991,53 @@ static void ggml_mul_mat_q4_1_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_1, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_1, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_1, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_1, VDR_Q4_1_Q8_1_MMQ, vec_dot_q4_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } + } } @@ -4050,17 +4045,52 @@ static void ggml_mul_mat_q5_0_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 128; + const int mmq_y = 64; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_0, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_0, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_0, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_0, VDR_Q5_0_Q8_1_MMQ, vec_dot_q5_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4068,17 +4098,52 @@ static void ggml_mul_mat_q5_1_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 128; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_1, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_1, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_1, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_1, VDR_Q5_1_Q8_1_MMQ, vec_dot_q5_1_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4086,17 +4151,52 @@ static void ggml_mul_mat_q8_0_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 128; + const int mmq_y = 64; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q8_0, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q8_0, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q8_0, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q8_0, VDR_Q8_0_Q8_1_MMQ, vec_dot_q8_0_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4104,17 +4204,52 @@ static void ggml_mul_mat_q2_K_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q2_K, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q2_K, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q2_K, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q2_K, VDR_Q2_K_Q8_1_MMQ, vec_dot_q2_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4122,17 +4257,52 @@ static void ggml_mul_mat_q3_K_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 128; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q3_K, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q3_K, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q3_K, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q3_K, VDR_Q3_K_Q8_1_MMQ, vec_dot_q3_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4140,17 +4310,52 @@ static void ggml_mul_mat_q4_K_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_K, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_K, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 32; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q4_K, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q4_K, VDR_Q4_K_Q8_1_MMQ, vec_dot_q4_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4158,17 +4363,52 @@ static void ggml_mul_mat_q5_K_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 128; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_K, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_K, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q5_K, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q5_K, VDR_Q5_K_Q8_1_MMQ, vec_dot_q5_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4176,17 +4416,52 @@ static void ggml_mul_mat_q6_K_q8_1_cuda( const void * vx, const void * vy, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_y, const int nrows_dst, cudaStream_t stream) { - const int block_num_x = (nrows_x + GGML_CUDA_MMQ_Y - 1) / GGML_CUDA_MMQ_Y; - const int block_num_y = (ncols_y + WARP_SIZE - 1) / WARP_SIZE; - const dim3 block_nums(block_num_x, block_num_y, 1); - const dim3 block_dims(WARP_SIZE, WARP_SIZE/4, 1); + int id; + CUDA_CHECK(cudaGetDevice(&id)); + const int compute_capability = g_compute_capabilities[id]; - if (nrows_x % GGML_CUDA_MMQ_Y == 0) { - mul_mat_q, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + if (compute_capability >= CC_TURING) { + const int mmq_x = 64; + const int mmq_y = 64; + const int nwarps = 4; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q6_K, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q6_K, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } else { - mul_mat_q, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> - <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + const int mmq_x = 32; + const int mmq_y = 64; + const int nwarps = 8; + + const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y; + const int block_num_y = (ncols_y + mmq_x - 1) / mmq_x; + const dim3 block_nums(block_num_x, block_num_y, 1); + const dim3 block_dims(WARP_SIZE, nwarps, 1); + + if (nrows_x % mmq_y == 0) { + const bool need_check = false; + mul_mat_q, + load_tiles_q6_K, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } else { + const bool need_check = true; + mul_mat_q, + load_tiles_q6_K, VDR_Q6_K_Q8_1_MMQ, vec_dot_q6_K_q8_1_mul_mat> + <<>>(vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y, nrows_dst); + } } } @@ -4361,20 +4636,6 @@ static void ggml_cuda_pool_free(void * ptr, size_t size) { } -static void * g_scratch_buffer = nullptr; -static size_t g_scratch_size = 1024*1024*1024; // 1 GB by default -static size_t g_scratch_offset = 0; - -static int g_device_count = -1; -static int g_main_device = 0; -static int g_compute_capabilities[GGML_CUDA_MAX_DEVICES]; -static float g_tensor_split[GGML_CUDA_MAX_DEVICES] = {0}; -static bool g_mul_mat_q = false; - -static cublasHandle_t g_cublas_handles[GGML_CUDA_MAX_DEVICES] = {nullptr}; - -static cudaStream_t g_cudaStreams_main[GGML_CUDA_MAX_DEVICES] = { nullptr }; - void ggml_init_cublas() { static bool initialized = false; @@ -4730,6 +4991,37 @@ inline void ggml_cuda_op_mul_mat_q( (void) i1; } +static int64_t get_row_rounding(ggml_type type) { + int max_compute_capability = INT_MIN; + for (int id = 0; id < g_device_count; ++id) { + if (max_compute_capability < g_compute_capabilities[id] + && g_tensor_split[id] < (id + 1 < g_device_count ? g_tensor_split[id + 1] : 1.0f)) { + max_compute_capability = g_compute_capabilities[id]; + } + } + + switch(type) { + case GGML_TYPE_Q4_0: + case GGML_TYPE_Q4_1: + return max_compute_capability >= CC_TURING ? 128 : 64; + case GGML_TYPE_Q5_0: + case GGML_TYPE_Q5_1: + case GGML_TYPE_Q8_0: + return 64; + case GGML_TYPE_F16: + return 1; + case GGML_TYPE_Q2_K: + case GGML_TYPE_Q3_K: + case GGML_TYPE_Q4_K: + case GGML_TYPE_Q5_K: + return max_compute_capability >= CC_TURING ? 128 : 64; + case GGML_TYPE_Q6_K: + return 64; + default: + GGML_ASSERT(false); + } +} + inline void ggml_cuda_op_mul_mat_vec( const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, char * src0_ddq_i, float * src0_ddf_i, float * src1_ddf_i, float * dst_ddf_i, int64_t i02, int64_t i01_low, int64_t i01_high, int i1, @@ -5130,14 +5422,16 @@ static void ggml_cuda_op(const ggml_tensor * src0, const ggml_tensor * src1, ggm int64_t row_low, row_high; if (split) { + const int64_t rounding = get_row_rounding(src0->type); + row_low = id == 0 ? 0 : nrows0*g_tensor_split[id]; - row_low -= row_low % GGML_CUDA_MMQ_Y; + row_low -= row_low % rounding; if (id == g_device_count - 1) { row_high = nrows0; } else { row_high = nrows0*g_tensor_split[id + 1]; - row_high -= row_high % GGML_CUDA_MMQ_Y; + row_high -= row_high % rounding; } } else { row_low = 0; @@ -5616,14 +5910,16 @@ void ggml_cuda_transform_tensor(void * data, struct ggml_tensor * tensor) { row_low = 0; row_high = nrows; } else if (backend == GGML_BACKEND_GPU_SPLIT) { + const int64_t rounding = get_row_rounding(tensor->type); + row_low = id == 0 ? 0 : nrows*g_tensor_split[id]; - row_low -= row_low % GGML_CUDA_MMQ_Y; + row_low -= row_low % rounding; if (id == g_device_count - 1) { row_high = nrows; } else { row_high = nrows*g_tensor_split[id + 1]; - row_high -= row_high % GGML_CUDA_MMQ_Y; + row_high -= row_high % rounding; } } else { GGML_ASSERT(false);