biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Kernel] Add per-tensor and per-token AZP epilogues (#5941)

Browse Source 
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
This commit is contained in:
Luka Govedič
2024-08-06 14:17:08 -04:00
committed by GitHub
parent 5c60c8c423
commit 8d59dbb000
11 changed files with 1175 additions and 153 deletions
Download Patch File Download Diff File Expand all files Collapse all files

111
csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu
View File

@@ -29,6 +29,40 @@ void cutlass_scaled_mm_sm90(torch::Tensor& c, torch::Tensor const& a,
c10::optional<torch::Tensor> const& bias);
#endif
void cutlass_scaled_mm_azp_sm75(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b,
torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
torch::Tensor const& azp_adj,
c10::optional<torch::Tensor> const& azp,
c10::optional<torch::Tensor> const& bias);
void cutlass_scaled_mm_azp_sm80(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b,
torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
torch::Tensor const& azp_adj,
c10::optional<torch::Tensor> const& azp,
c10::optional<torch::Tensor> const& bias);
void cutlass_scaled_mm_azp_sm89(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b,
torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
torch::Tensor const& azp_adj,
c10::optional<torch::Tensor> const& azp,
c10::optional<torch::Tensor> const& bias);
#if defined CUDA_VERSION && CUDA_VERSION >= 12000
void cutlass_scaled_mm_azp_sm90(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b,
torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
torch::Tensor const& azp_adj,
c10::optional<torch::Tensor> const& azp,
c10::optional<torch::Tensor> const& bias);
#endif
bool cutlass_scaled_mm_supports_fp8(int64_t cuda_device_capability) {
// CUTLASS FP8 kernels need at least
// CUDA 12.0 on SM90 systems (Hopper)
@@ -45,18 +79,20 @@ bool cutlass_scaled_mm_supports_fp8(int64_t cuda_device_capability) {
return false;
}
void cutlass_scaled_mm(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b, torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
c10::optional<torch::Tensor> const& bias) {
int32_t major_capability;
int32_t minor_capability;
int32_t get_sm_version_num() {
int32_t major_capability, minor_capability;
cudaDeviceGetAttribute(&major_capability, cudaDevAttrComputeCapabilityMajor,
0);
cudaDeviceGetAttribute(&minor_capability, cudaDevAttrComputeCapabilityMinor,
0);
int32_t version_num = major_capability * 10 + minor_capability;
return version_num;
}
void cutlass_scaled_mm(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b, torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
c10::optional<torch::Tensor> const& bias) {
// Checks for conformality
TORCH_CHECK(a.dim() == 2 && b.dim() == 2 && c.dim() == 2);
TORCH_CHECK(c.size(0) == a.size(0) && a.size(1) == b.size(0) &&
@@ -77,7 +113,7 @@ void cutlass_scaled_mm(torch::Tensor& c, torch::Tensor const& a,
}
at::cuda::OptionalCUDAGuard const device_guard(device_of(a));
int32_t version_num = get_sm_version_num();
if (version_num >= 90) {
// Hopper
@@ -99,3 +135,64 @@ void cutlass_scaled_mm(torch::Tensor& c, torch::Tensor const& a,
cutlass_scaled_mm_sm75(c, a, b, a_scales, b_scales, bias);
}
}
void cutlass_scaled_mm_azp(torch::Tensor& c, torch::Tensor const& a,
torch::Tensor const& b,
torch::Tensor const& a_scales,
torch::Tensor const& b_scales,
torch::Tensor const& azp_adj,
c10::optional<torch::Tensor> const& azp,
c10::optional<torch::Tensor> const& bias) {
// Checks for conformality
TORCH_CHECK(a.dim() == 2 && b.dim() == 2 && c.dim() == 2);
TORCH_CHECK(c.size(0) == a.size(0) && a.size(1) == b.size(0) &&
b.size(1) == c.size(1));
TORCH_CHECK(a_scales.numel() == 1 || a_scales.numel() == a.size(0));
TORCH_CHECK(b_scales.numel() == 1 || b_scales.numel() == b.size(1));
// Check for strides and alignment
TORCH_CHECK(a.stride(1) == 1 && c.stride(1) == 1); // Row-major
TORCH_CHECK(b.stride(0) == 1); // Column-major
TORCH_CHECK(c.stride(0) % 16 == 0 &&
b.stride(1) % 16 == 0); // 16 Byte Alignment
TORCH_CHECK(a_scales.is_contiguous() && b_scales.is_contiguous());
// bias, azp, azp_adj are all 1d
// bias and azp_adj have n elements, azp has m elements
if (bias) {
TORCH_CHECK(bias->numel() == b.size(1) && bias->is_contiguous());
}
if (azp) {
TORCH_CHECK(azp->numel() == a.size(0) && azp->is_contiguous());
}
TORCH_CHECK(azp_adj.numel() == b.size(1) && azp_adj.is_contiguous());
// azp & bias types
TORCH_CHECK(azp_adj.dtype() == torch::kInt32);
TORCH_CHECK(!azp || azp->dtype() == torch::kInt32);
TORCH_CHECK(!bias || bias->dtype() == c.dtype(),
"currently bias dtype must match output dtype ", c.dtype());
at::cuda::OptionalCUDAGuard const device_guard(device_of(a));
int32_t version_num = get_sm_version_num();
if (version_num >= 90) {
// Hopper
// Guard against compilation issues for sm90 kernels
#if defined CUDA_VERSION && CUDA_VERSION >= 12000
cutlass_scaled_mm_azp_sm90(c, a, b, a_scales, b_scales, azp_adj, azp, bias);
#else
cutlass_scaled_mm_azp_sm80(c, a, b, a_scales, b_scales, azp_adj, azp, bias);
#endif
} else if (version_num == 89) {
// Ada Lovelace
cutlass_scaled_mm_azp_sm89(c, a, b, a_scales, b_scales, azp_adj, azp, bias);
} else if (version_num >= 80) {
// Ampere
cutlass_scaled_mm_azp_sm80(c, a, b, a_scales, b_scales, azp_adj, azp, bias);
} else {
// Turing
TORCH_CHECK(version_num >= 75);
cutlass_scaled_mm_azp_sm75(c, a, b, a_scales, b_scales, azp_adj, azp, bias);
}
}