Skip to content

Migrate GenAI quantize kernels to FBGEMM_LAUNCH_KERNEL, pt 4 #4863

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Migrate GenAI quantize kernels to FBGEMM_LAUNCH_KERNEL, pt 4 #4863

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
50 changes: 39 additions & 11 deletions fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1214,9 +1214,16 @@ void invokeComputeScalesAndQuantizeMatrixCol(
dim3 grid((lda + CTA_SIZE - 1) / CTA_SIZE);
C10_CUDA_CHECK(cudaMemsetAsync(quant_ptr, 0, lda * sizeof(T_S), stream));
C10_CUDA_KERNEL_LAUNCH_CHECK();
computeFP8QuantizeScaleColwise<<<grid, block, 0, stream>>>(
quant_ptr, input, numel, lda);
C10_CUDA_KERNEL_LAUNCH_CHECK();
FBGEMM_LAUNCH_KERNEL(
(computeFP8QuantizeScaleColwise<T_S, T_IN>),
grid,
block,
0,
stream,
quant_ptr,
input,
numel,
lda);
invokeQuantizeMatrixColwise(output, quant_ptr, input, numel, lda, stream);
}

Expand Down Expand Up @@ -1639,15 +1646,25 @@ void invokeFP4Quantization(

// Launch the cvt kernel.
if (useUE8M0) {
cvt_fp16_to_fp4<T, true><<<grid, block, 0, stream>>>(
FBGEMM_LAUNCH_KERNEL(
(cvt_fp16_to_fp4<T, true>),
grid,
block,
0,
stream,
m,
n,
input,
SFScale,
reinterpret_cast<uint32_t*>(output),
reinterpret_cast<uint32_t*>(SFOuput));
} else {
cvt_fp16_to_fp4<T, false><<<grid, block, 0, stream>>>(
FBGEMM_LAUNCH_KERNEL(
(cvt_fp16_to_fp4<T, false>),
grid,
block,
0,
stream,
m,
n,
input,
Expand Down Expand Up @@ -1924,10 +1941,17 @@ void fp4_fused_amax_quantize(
const dim3 block(blocksize, blocks_per_cta);
const int blocks = ceil_div(numel, blocksize * blocks_per_cta);

compute_amax_and_quantize_kernel<__nv_bfloat16, 16, 4>
<<<blocks, block, 0, stream>>>(x, y, numel, blocksize, global_amax_ptr);

C10_CUDA_KERNEL_LAUNCH_CHECK();
FBGEMM_LAUNCH_KERNEL(
(compute_amax_and_quantize_kernel<__nv_bfloat16, 16, 4>),
blocks,
block,
0,
stream,
x,
y,
numel,
blocksize,
global_amax_ptr);
}

template <typename T_S, typename T_W>
Expand Down Expand Up @@ -1974,15 +1998,19 @@ void invokeComputeFP4GlobalAmax(
constexpr dim3 grid(1024);
int64_t numel_scale = numel;
C10_CUDA_CHECK(cudaMemsetAsync(quant_ptr, 0, sizeof(T_S), stream));
computeFP4GlobalAmax<<<grid, block, 0, stream>>>(
FBGEMM_LAUNCH_KERNEL(
(computeFP4GlobalAmax<T_S, T_IN>),
grid,
block,
0,
stream,
quant_ptr,
input,
numel_scale,
lda,
total_elements_per_slice,
bs,
scale_ub);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}

std::vector<at::Tensor> fake_quantize_nvfp4_per_tensor(
Expand Down
Loading