[AMD] Reorganize hip-related header files in sgl-kernel (#9320)

This commit is contained in:
Hubert Lu
2025-08-18 16:53:44 -07:00
committed by GitHub
parent c2fbf60f39
commit c6c379ab31
14 changed files with 18 additions and 17 deletions

View File

@@ -67,7 +67,7 @@ __global__ void per_token_quant_fp8_kernel(
for (uint32_t j = 0; j < kVecSize; ++j) {
float val = static_cast<float>(input_vec[j]) * scale_inv;
val = fmaxf(fminf(val, FP8_E4M3_MAX), -FP8_E4M3_MAX);
#ifndef USE_ROCM
#if !defined(USE_ROCM) || defined(HIP_FP8_TYPE_E4M3)
output_arr[j] = static_cast<DST_DTYPE>(val);
#else
output_arr[j] = c10::Float8_e4m3fnuz(
@@ -143,7 +143,7 @@ __global__ void per_token_quant_fp8_small_batch_kernel(
#pragma unroll
for (uint32_t j = 0; j < kVecSize; ++j) {
float val = fmaxf(fminf(static_cast<float>(input_vec[j]) * scale_inv, FP8_E4M3_MAX), -FP8_E4M3_MAX);
#ifndef USE_ROCM
#if !defined(USE_ROCM) || defined(HIP_FP8_TYPE_E4M3)
output_arr[j] = static_cast<DST_DTYPE>(val);
#else
output_arr[j] = c10::Float8_e4m3fnuz(