ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-17 17:19:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

working prefill shapes

Browse Source
This commit is contained in:
khushbu
2025-12-16 18:21:44 -05:00
parent 5a3e7de060
commit 373d89d381
4 changed files with 63 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
example/ck_tile/38_block_scale_gemm/gemm_bquant_quantgrouped_preshufflequant.cpp
View File

@@ -4,7 +4,8 @@
#include "run_gemm_quant_example.inc"
template <typename T>
using GemmConfig = GemmConfigPreshuffleBQuantPrefill<T>;
using GemmConfig = GemmConfigPreshuffleBQuantPrefill<T>; // GemmConfigPreshuffleQuantDecode<T>;
// //GemmConfigPreshuffleBQuantPrefill<T>;
void bquant_quantgrouped_preshufflequant_instance_factory(
std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut)

6
example/ck_tile/38_block_scale_gemm/gemm_quant.cpp
View File

@@ -33,14 +33,14 @@ auto create_args(int argc, char* argv[])
"fp8",
"Data type. For AQuant: fp8, bf8, i4fp8, or i4bf8; for Bquant: fp8, bf8, fp8i4, "
"bf8i4 or bf16fp4")
.insert("warmup", "50", "Number of iterations before benchmarking the kernel")
.insert("repeat", "1000", "Number of iterations to benchmark the kernel")
.insert("warmup", "1", "Number of iterations before benchmarking the kernel")
.insert("repeat", "0", "Number of iterations to benchmark the kernel")
.insert("timer", "gpu", "gpu:gpu timer, cpu:cpu timer")
.insert("split_k", "1", "SplitK value")
.insert("device", "0", "Device id that will be used to run the kernel")
.insert("init", "0", "0:random, 1:linear, 2:constant(1)")
.insert("flush_cache", "true", "Flush cache before running the kernel")
.insert("rotating_count", "1000", "Rotating count")
.insert("rotating_count", "0", "Rotating count")
.insert("quant_mode", "bquant", "Choose aquant, bquant, tensor or rowcol")
.insert("preshuffleb", "false", "Enable preshuffle of tensor B")
.insert("preshufflequant", "false", "Enable preshuffle of quant tensor")