EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Speed up per token and per tensor quant by 15% (#4639)

Browse Source
This commit is contained in:
Chunan Zeng
2025-03-22 00:37:57 -07:00
committed by GitHub
parent f69e0696bc
commit 6a384d5c01
2 changed files with 29 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
sgl-kernel/csrc/gemm/per_tensor_quant_fp8.cu
View File

@@ -54,42 +54,40 @@ __global__ void per_tensor_quant_fp8_kernel(
const int grid_size = blockDim.x * gridDim.x; const int grid_size = blockDim.x * gridDim.x;
const float scale_val = 1.0f / (*scale); const float scale_val = 1.0f / (*scale);
constexpr uint32_t vec_size = 16 / sizeof(T); // We want to store 128 bits of data at a time. 16 = 128 / 8 bits
using vec_t = flashinfer::vec_t<T, vec_size>; // Load is already vectorized, so 16 elements work for T.
const uint32_t VEC_SIZE = 16;
using vec_t = flashinfer::vec_t<T, VEC_SIZE>;
const int32_t num_vec_elems = num_elements / vec_size; const int32_t num_vec_elems = num_elements / VEC_SIZE;
for (int32_t i = gid; i < num_vec_elems; i += grid_size) { for (int32_t i = gid; i < num_vec_elems; i += grid_size) {
vec_t input_vec; vec_t input_vec;
input_vec.cast_load(input + i * vec_size); input_vec.cast_load(input + i * VEC_SIZE);
FP8_TYPE output_arr[vec_size]; FP8_TYPE output_arr[VEC_SIZE];
#pragma unroll #pragma unroll
for (uint32_t j = 0; j < vec_size; ++j) { for (uint32_t j = 0; j < VEC_SIZE; ++j) {
float val = fmax(fmin(static_cast<float>(input_vec[j]) * scale_val, FP8_E4M3_MAX), -FP8_E4M3_MAX); float val = fmax(fmin(static_cast<float>(input_vec[j]) * scale_val, FP8_E4M3_MAX), -FP8_E4M3_MAX);
#ifndef USE_ROCM #ifndef USE_ROCM
output_arr[j] = static_cast<FP8_TYPE>(val); output_arr[j] = static_cast<FP8_TYPE>(val);
#else #else
output_arr[j] = c10::Float8_e4m3fnuz( output_arr[j] = c10::Float8_e4m3fnuz(
__hip_cvt_float_to_fp8(value, fp8::fp8_type::__default_saturation, fp8::fp8_type::__default_interpret), __hip_cvt_float_to_fp8(val, fp8::fp8_type::__default_saturation, fp8::fp8_type::__default_interpret),
c10::Float8_e4m3fnuz::from_bits()); c10::Float8_e4m3fnuz::from_bits());
#endif #endif
} }
*(uint4*)(output + i * VEC_SIZE) = *(uint4*)output_arr;
#pragma unroll
for (uint32_t j = 0; j < vec_size; ++j) {
output[i * vec_size + j] = output_arr[j];
}
} }
const int32_t remaining_start = num_vec_elems * vec_size; const int32_t remaining_start = num_vec_elems * VEC_SIZE;
for (int32_t idx = remaining_start + gid; idx < num_elements; idx += grid_size) { for (int32_t idx = remaining_start + gid; idx < num_elements; idx += grid_size) {
float val = fmax(-FP8_E4M3_MAX, fmin(static_cast<float>(input[idx]) * scale_val, FP8_E4M3_MAX)); float val = fmax(-FP8_E4M3_MAX, fmin(static_cast<float>(input[idx]) * scale_val, FP8_E4M3_MAX));
#ifndef USE_ROCM #ifndef USE_ROCM
output[idx] = static_cast<FP8_TYPE>(val); output[idx] = static_cast<FP8_TYPE>(val);
#else #else
output[idx] = c10::Float8_e4m3fnuz( output[idx] = c10::Float8_e4m3fnuz(
__hip_cvt_float_to_fp8(value, fp8::fp8_type::__default_saturation, fp8::fp8_type::__default_interpret), __hip_cvt_float_to_fp8(val, fp8::fp8_type::__default_saturation, fp8::fp8_type::__default_interpret),
c10::Float8_e4m3fnuz::from_bits()); c10::Float8_e4m3fnuz::from_bits());
#endif #endif
} }

35
sgl-kernel/csrc/gemm/per_token_quant_fp8.cu
View File

@@ -24,17 +24,19 @@ __global__ void per_token_quant_fp8_kernel(
float max_value = 0.0f; float max_value = 0.0f;
constexpr uint32_t vec_size = 16 / sizeof(T); // We want to store 128 bits of data at a time. 16 = 128 / 8 bits
using vec_t = flashinfer::vec_t<T, vec_size>; // Load is already vectorized, so 16 elements work for T.
const int32_t num_vec_elems = hidden_dim / vec_size; const uint32_t VEC_SIZE = 16;
using vec_t = flashinfer::vec_t<T, VEC_SIZE>;
const int32_t num_vec_elems = hidden_dim / VEC_SIZE;
// Find max using vectorized loads // Find max using vectorized loads
for (int32_t i = tid; i < num_vec_elems; i += block_dim) { for (int32_t i = tid; i < num_vec_elems; i += block_dim) {
vec_t input_vec; vec_t input_vec;
input_vec.cast_load(token_input + i * vec_size); input_vec.cast_load(token_input + i * VEC_SIZE);
#pragma unroll #pragma unroll
for (uint32_t j = 0; j < vec_size; ++j) { for (uint32_t j = 0; j < VEC_SIZE; ++j) {
float val = static_cast<float>(input_vec[j]); float val = static_cast<float>(input_vec[j]);
max_value = fmaxf(max_value, fabsf(val)); max_value = fmaxf(max_value, fabsf(val));
} }
@@ -42,24 +44,24 @@ __global__ void per_token_quant_fp8_kernel(
max_value = blockReduceMax(max_value); max_value = blockReduceMax(max_value);
__shared__ float block_max; __shared__ float scale;
if (tid == 0) { if (tid == 0) {
block_max = max_value / FP8_E4M3_MAX; scale = max_value / FP8_E4M3_MAX;
output_s[token_idx] = block_max; output_s[token_idx] = scale;
} }
__syncthreads(); __syncthreads();
const float scale_val = 1.0f / block_max; const float scale_inv = 1.0f / scale;
// Quantize using vectorized loads // Quantize using vectorized loads
for (int32_t i = tid; i < num_vec_elems; i += block_dim) { for (int32_t i = tid; i < num_vec_elems; i += block_dim) {
vec_t input_vec; vec_t input_vec;
input_vec.cast_load(token_input + i * vec_size); input_vec.cast_load(token_input + i * VEC_SIZE);
FP8_TYPE output_arr[vec_size]; FP8_TYPE output_arr[VEC_SIZE];
#pragma unroll #pragma unroll
for (uint32_t j = 0; j < vec_size; ++j) { for (uint32_t j = 0; j < VEC_SIZE; ++j) {
float val = fmaxf(fminf(static_cast<float>(input_vec[j]) * scale_val, FP8_E4M3_MAX), -FP8_E4M3_MAX); float val = fmaxf(fminf(static_cast<float>(input_vec[j]) * scale_inv, FP8_E4M3_MAX), -FP8_E4M3_MAX);
#ifndef USE_ROCM #ifndef USE_ROCM
output_arr[j] = static_cast<FP8_TYPE>(val); output_arr[j] = static_cast<FP8_TYPE>(val);
#else #else
@@ -69,10 +71,7 @@ __global__ void per_token_quant_fp8_kernel(
#endif #endif
} }
#pragma unroll *(uint4*)(token_output + i * VEC_SIZE) = *(uint4*)output_arr;
for (uint32_t j = 0; j < vec_size; ++j) {
token_output[i * vec_size + j] = output_arr[j];
}
} }
} }
@@ -85,7 +84,7 @@ void sgl_per_token_quant_fp8(torch::Tensor input, torch::Tensor output_q, torch:
const int64_t num_tokens = input_sizes[0]; const int64_t num_tokens = input_sizes[0];
const int64_t hidden_dim = input_sizes[1]; const int64_t hidden_dim = input_sizes[1];
TORCH_CHECK(hidden_dim % 8 == 0, "Hidden dimension must be divisible by 8, but got ", hidden_dim); TORCH_CHECK(hidden_dim % 16 == 0, "Hidden dimension must be divisible by 16, but got ", hidden_dim);
const int block_size = 256; const int block_size = 256;
const int num_blocks = num_tokens; const int num_blocks = num_tokens;