louis_lifu
/
vllm
mirror of https://github.com/vllm-project/vllm
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

workaround of AWQ for Turing GPUs (#1252)

This commit is contained in:
twaka 2023-10-11 11:48:16 +09:00 committed by GitHub
parent 91fce82c6f
commit 8285736840
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 73 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/quantization/awq/dequantize.cuh
View File

@ -16,7 +16,7 @@ namespace awq {
__device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source)
{
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
assert(false);
#else
uint4 result;

72
csrc/quantization/awq/gemm_kernels.cu
View File

@ -29,7 +29,7 @@ __pack_half2(const half x, const half y) {
__global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n128k32(int G, int split_k_iters, half* __restrict__ A, int* __restrict__ B, half* __restrict__ scaling_factors, int* __restrict__ zeros, int M, int IC, int OC, half* __restrict__ C)
{
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
assert(false);
#else
static constexpr uint32_t ZERO = 0x0;
@ -191,6 +191,39 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n128k32(int G, i
}
}
for (int j_0_4 = 0; j_0_4 < 4; ++j_0_4) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ == 750
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
#else
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32"
@ -206,6 +239,8 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n128k32(int G, i
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
#endif
}
}
}
@ -226,7 +261,7 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n128k32(int G, i
__global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n64k32(int G, int split_k_iters, half* __restrict__ A, int* __restrict__ B, half* __restrict__ scaling_factors, int* __restrict__ zeros, int M, int IC, int OC, half* __restrict__ C)
{
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
assert(false);
#else
static constexpr uint32_t ZERO = 0x0;
@ -392,7 +427,39 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n64k32(int G, in
for (int j_0_4 = 0; j_0_4 < 2; ++j_0_4)
{
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ == 750
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
}
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
"{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
#else
{
__asm__ __volatile__(
"mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32"
@ -408,6 +475,7 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16n64k32(int G, in
: "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
: "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
}
#endif
}
}
}

4
vllm/model_executor/quantization_utils/awq.py
View File

@ -42,8 +42,8 @@ class AWQConfig(QuantizationConfig):
@classmethod
def get_min_capability(cls) -> int:
# The AWQ kernel only supports Ampere or newer GPUs.
return 80
# The AWQ kernel only supports Turing or newer GPUs.
return 75
@classmethod
def get_config_filenames(cls) -> List[str]: