merge from public repo

09d4c3a4 · illsilin · 171ed358 · 8e4c3fb1 · 09d4c3a4 · 09d4c3a4
Commit 09d4c3a4 authored Oct 01, 2024 by illsilin
20 changed files
--- a/include/ck_tile/host/host_tensor.hpp
+++ b/include/ck_tile/host/host_tensor.hpp
--- a/include/ck_tile/host/reference/reference_gemm.hpp
+++ b/include/ck_tile/host/reference/reference_gemm.hpp
--- a/include/ck_tile/host/reference/reference_im2col.hpp
+++ b/include/ck_tile/host/reference/reference_im2col.hpp
--- a/include/ck_tile/ops/fmha/block/block_masking.hpp
+++ b/include/ck_tile/ops/fmha/block/block_masking.hpp
--- a/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_combine_kernel.hpp
+++ b/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_combine_kernel.hpp
--- a/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_combine_tile_partitioner.hpp
+++ b/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_combine_tile_partitioner.hpp
--- a/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_kernel.hpp
+++ b/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_kernel.hpp
--- a/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_tile_partitioner.hpp
+++ b/include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_tile_partitioner.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_bwd_dq_dk_dv_pipeline_kr_ktr_vr_iglp.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_bwd_dq_dk_dv_pipeline_kr_ktr_vr_iglp.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_bwd_pipeline_default_policy.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_bwd_pipeline_default_policy.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_splitkv_combine_pipeline.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_splitkv_combine_pipeline.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_splitkv_pipeline_qr_ks_vs.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_splitkv_pipeline_qr_ks_vs.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs.hpp
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs_async.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs_async.hpp
@@ -268,7 +268,7 @@ struct BlockFmhaPipelineQRKSVSAsync
        const auto num_total_loop = integer_divide_ceil(seqlen_k_end - seqlen_k_start, kN0);
-        // check early exit
+        // check early exit if no work to do
        if constexpr(FmhaMask::IsMasking || kPadSeqLenK)
        {
            if(num_total_loop <= 0)

--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qx_ks_vs_custom_policy.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qx_ks_vs_custom_policy.hpp
--- a/include/ck_tile/ops/gemm.hpp
+++ b/include/ck_tile/ops/gemm.hpp
--- a/include/ck_tile/ops/gemm/block/block_gemm_areg_bgmem_creg_v1.hpp
+++ b/include/ck_tile/ops/gemm/block/block_gemm_areg_bgmem_creg_v1.hpp
--- a/include/ck_tile/ops/gemm/block/block_gemm_asmem_bsmem_creg_v1_default_policy.hpp
+++ b/include/ck_tile/ops/gemm/block/block_gemm_asmem_bsmem_creg_v1_default_policy.hpp
--- a/include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
+++ b/include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
--- a/include/ck_tile/ops/gemm/kernel/gemm_tile_partitioner.hpp
+++ b/include/ck_tile/ops/gemm/kernel/gemm_tile_partitioner.hpp