Merge pull request #1 from aska-0096/e2e_v2

Merging some forgot changes

Merge pull request #1 from aska-0096/e2e_v2
Merging some forgot changes
a29d2b76 · Haocong WANG · GitHub · 2c265ebd · 87ae9b71 · a29d2b76
Unverified Commit a29d2b76 authored Apr 19, 2023 by Haocong WANG Committed by GitHub Apr 19, 2023
Showing with 4 additions and 4 deletions

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp ...emm/batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp +3 -3

include/ck/ck.hpp include/ck/ck.hpp +1 -1

No files found.
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp
@@ -96,12 +96,12 @@ using DeviceGemmInstance =
        256,
        //      Gemm 0
        128, // MPerBlock
-        64, // LPerBlock
-        32,  // KPerBlock
+        64,  // LPerBlock
+        64,  // KPerBlock
        8,   // K1
        //      Gemm 1
        64, // NPerBlock
-        32, // LTilePerBlock
+        64, // LTilePerBlock
        8,  // L1
        16, // MPerWMMA
        16, // LPerWMMA

--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -36,7 +36,7 @@
 #elif defined(__gfx1030__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
 #elif defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) // for GPU code
-#define CK_BUFFER_RESOURCE_3RD_DWORD 0x10020000
+#define CK_BUFFER_RESOURCE_3RD_DWORD 0x31004000
 #endif

 // FMA instruction