From dc54bb272f26a07f23e11dcc1f260eae87eef171 Mon Sep 17 00:00:00 2001
From: Thomas Ning <Thomas.Ning@amd.com>
Date: Thu, 8 May 2025 00:07:03 -0700
Subject: [PATCH] Revert "Disable the SMFMA instruction for gfx90a. (#2174)"
 (#2175)

This reverts commit ad29f1b98abd7096d447a095a705ca0577b875d9.

[ROCm/composable_kernel commit: c757046d49e5e5bbd3b3c9bfda95cd093e70f0e8]
---
 include/ck_tile/ops/gemm/warp/warp_gemm.hpp   | 3 +--
 tile_engine/ops/gemm/gemm_instance_builder.py | 6 +-----
 2 files changed, 2 insertions(+), 7 deletions(-)

diff --git a/include/ck_tile/ops/gemm/warp/warp_gemm.hpp b/include/ck_tile/ops/gemm/warp/warp_gemm.hpp
index 61c61c2d9a..c98d46e3a0 100644
--- a/include/ck_tile/ops/gemm/warp/warp_gemm.hpp
+++ b/include/ck_tile/ops/gemm/warp/warp_gemm.hpp
@@ -97,13 +97,12 @@ using WarpGemmMfmaF16F16F32M64N4K16 = WarpGemmImpl<WarpGemmAtrributeMfmaIterateK
     4>>;
 
 // fp16 2:4 structured sparsity
-#if defined(__gfx94__) || defined(__gfx950__)
+
 using WarpGemmSmfmacF16F16F32M32N32K16 = WarpGemmSmfmacImpl<WarpGemmAttributeSmfmac<
     WarpGemmAttributeSmfmacImplF16F16F32M32N32K16<WGAttrCtlEnum::Default_>>>;
 
 using WarpGemmSmfmacF16F16F32M16N16K32 = WarpGemmSmfmacImpl<WarpGemmAttributeSmfmac<
     WarpGemmAttributeSmfmacImplF16F16F32M16N16K32<WGAttrCtlEnum::Default_>>>;
-#endif
 
 // bf16
 using WarpGemmMfmaBf16Bf16F32M32N32K8 = WarpGemmImpl<
diff --git a/tile_engine/ops/gemm/gemm_instance_builder.py b/tile_engine/ops/gemm/gemm_instance_builder.py
index c00554df8f..3839523e3d 100755
--- a/tile_engine/ops/gemm/gemm_instance_builder.py
+++ b/tile_engine/ops/gemm/gemm_instance_builder.py
@@ -535,11 +535,7 @@ struct GemmDispatcher {
                     ((tile[6] == 32 and tile[7] == 32 and tile[8] == 16) or
                      (tile[6] == 16 and tile[7] == 16 and tile[8] == 32))
                 content += f"""
-#if defined(__gfx908__)
-                run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(False)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);
-#else
-                run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(sparse)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);
-#endif"""
+                run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(sparse)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);"""
             content += f"""
             }} else {{"""
             for tile in tile_params: