Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Programmatic Dependent Launch (PDL) for more performance on newer NVIDIA GPUs (Hopper+) (#22522)

Browse Source 
* Adds initial PDL setup.

* Adds PDL barriers based on simple heuristic: place "sync" before first input pointer access, and "launch" after last write, e.g. to tensors like dst.

* Further optimization pass of the first half of kernels

* Optimized PDL barriers for the second batch of kernels

* Further refinements after rebase.

* Moves pdl logic to separate function, removes some whitespace

* Strips post-hoc PDL logic

* Adds stream capture PDL setup. Enrolls quantize_q8_1 to leverage pdl to
overlap execution with previous kernels

* Enrolls mul_mat_vec_q, rms_norm_f32 and k_bin_bcast (partly) into PDL

* Enrolls mmvf, rope, set-rows and topk kernels for gpt-oss into PDL

* Introduce ggml_cuda_kernel_launch, to abstract away cudaLaunchKernelEx,
to enable hip/musa compatibility

* Enrolls cpy_scalar_contiguous, k_get_rows_float and rms_norm_f32

* Enrolls flash_attn_combine_results

* Fix: Drops needless and broken check of CUDA arch for PDL. PDL either
works or is without effect.

* Enrolls flash-attention kernels to pdl

* Fix: inlines ggml_cuda_kernel_launch, and uses perfect forwarding for
kernels args. This fixes PDL.

* Perf: Enrolls k_bin_bcast variadic template invocation into PDL, via
and template alias and template expansion

* Enrolls all remaining kernels for qwen3-coder-next into PDL

* Remove all PDL LC calls to create a baseline

* Added LC according to internal guidance and tested kernel performance.

* Enrols missing qwen3-5 kernels passively into PDL.

* Kernel optimizations (LC signals) for qwen3.5

* Enrolls ssm-scan kernels into PDL

* Adds GGML_CUDA_PDL command line option to toggle PDL.

* Fix: Ada and lower compilation by guarding PDL calls correctly

* Cleanup: Removes commented out GGML_CUDA_PDL_LC

* Cleanup: Removes experimental comments

* Adds 90-virtual to build script so that Hopper GPUs can leverage PDL.

* Adds stricter checks to enable PDL, adds env-check to disable it, and removes now superfluous compile option to enable PDL.

* Fix: Correct PDL en/disablement based on device-side arch check. Host
side check is UB. Required moving from macros to inlined functions

* Fix: default-disable PDL. Enable by setting GGML_CUDA_ENABLE_PDL=1

* Enable PDL by default for Hopper+ devices

* Enrolls softcap_f32 and two flash_attn kernels into PDL.

* Improves flash attn PDL barrier placement

* Fix: Perf regression on ada; excludes ada and below from PDL launches

* Improves some sync barrier placements

* Drops superfluous constructor

* Adds #endif guard comments

* Reverts experimental change to top-k-moe.cu, which moved expensive allocations
in front of the PDL barrier. It did not have a meaningful impact.

* Exchanges GGML_CUDA_DISABLE_PDL with GGML_CUDA_PDL. IFF GGML_CUDA_PDL=0
PDL is disabled

* Revert "Drops superfluous constructor". Adds const to remaining
arguments

This reverts commit 12b1d250da0089ae02a9bb71bbb3fd6d70f6f2f1.

* Cleanup: Removes and fixes some comments and whitespace

* Clarifies comment of sync-barrier position

* Relocates and refactors PDL launch functions and accessories

* Adds error checking to the regular kernel launch path

* Drops "auto" in favor of "ggml_cuda_kernel_params"

* Adds "const" to ggml_cuda_kernel_launch_params

* [Whitespace] Adds final newline to common.cuh to make editorconfig CI job happy
This commit is contained in:
Andreas Kieslinger
2026-05-20 13:59:02 +02:00
committed by GitHub
parent 29f1482221
commit e947228222
27 changed files with 310 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/topk-moe.cu
+25 -22
View File
@@ -105,6 +105,7 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
wt[i] = -INFINITY;
}
ggml_cuda_pdl_sync();
#pragma unroll
for (int i = 0; i < n_experts; i += WARP_SIZE) {
const int expert = i + threadIdx.x;
@@ -161,6 +162,7 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
output_weights[i] = 0.f;
}
ggml_cuda_pdl_lc();
for (int k = 0; k < n_expert_used; k++) {
float max_val = wt[0];
int max_expert = threadIdx.x;
@@ -271,51 +273,52 @@ static void launch_topk_moe_cuda(ggml_backend_cuda_context & ctx,
dim3 grid_dims((n_rows + rows_per_block - 1) / rows_per_block, 1, 1);
dim3 block_dims(WARP_SIZE, rows_per_block, 1);
cudaStream_t stream = ctx.stream();
const ggml_cuda_kernel_launch_params launch_params = ggml_cuda_kernel_launch_params(grid_dims, block_dims, 0, stream);
switch (n_expert) {
case 1:
topk_moe_cuda<1, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<1, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 2:
topk_moe_cuda<2, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<2, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 4:
topk_moe_cuda<4, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<4, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 8:
topk_moe_cuda<8, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<8, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 16:
topk_moe_cuda<16, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<16, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 32:
topk_moe_cuda<32, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<32, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 64:
topk_moe_cuda<64, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<64, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 128:
topk_moe_cuda<128, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<128, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 256:
topk_moe_cuda<256, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<256, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 512:
topk_moe_cuda<512, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<512, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 576:
topk_moe_cuda<576, has_bias><<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, bias, n_rows, n_expert_used,
clamp_val, scale_val, config);
ggml_cuda_kernel_launch(topk_moe_cuda<576, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
default:
GGML_ASSERT(false && "fatal error");