Merge remote-tracking branch 'origin/refactor/grp-hadamard-ldmatrix-transpose' into refactor/grp-hadamard-ldmatrix-transpose

cael-ling · cael-ling · commit d90ac010e4cf · 2026-03-29T02:52:28.000-07:00
Made-with: Cursor
diff --git a/transformer_engine/common/hadamard_transform/group_hadamard_transform.cu b/transformer_engine/common/hadamard_transform/group_hadamard_transform.cu
@@ -90,8 +90,7 @@ __device__ __forceinline__ void ComputeKernel(uint32_t b_frag_i[4], uint32_t b_f
   }
 
   if (kReturnIdentityAmax) {
-    if (kReturnTransposedAmax ||
-        (!kReturnTransposedAmax && !kReturnPreRhtAmax)) {
+    if (kReturnTransposedAmax || (!kReturnTransposedAmax && !kReturnPreRhtAmax)) {
       ldmatrix_x4_m8n8_shared_b16<false>(a_frag[0], a_frag[1], a_frag[2], a_frag[3],
                                          reinterpret_cast<uint4*>(in_sh_ptr) + swizzle_idx);
     }

Original file line number	Diff line number	Diff line change
`@@ -90,8 +90,7 @@ __device__ __forceinline__ void ComputeKernel(uint32_t b_frag_i[4], uint32_t b_f`
`90`	`90`	`}`
`91`	`91`
`92`	`92`	`if (kReturnIdentityAmax) {`
`93`		`- if (kReturnTransposedAmax \|\|`
`94`		`- (!kReturnTransposedAmax && !kReturnPreRhtAmax)) {`
	`93`	`+ if (kReturnTransposedAmax \|\| (!kReturnTransposedAmax && !kReturnPreRhtAmax)) {`
`95`	`94`	`ldmatrix_x4_m8n8_shared_b16<false>(a_frag[0], a_frag[1], a_frag[2], a_frag[3],`
`96`	`95`	`reinterpret_cast<uint4*>(in_sh_ptr) + swizzle_idx);`
`97`	`96`	`}`