add 2:4 sparsity matrix ops to rocdl

Author: SamGinzburg

mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td

@@ -408,6 +408,24 @@ def ROCDL_mfma_i32_32x32x32_i8 : ROCDL_Mfma_IntrOp<"mfma.i32.32x32x32.i8">;
 def ROCDL_mfma_f32_32x32x16_f16 : ROCDL_Mfma_IntrOp<"mfma.f32.32x32x16.f16">;
 def ROCDL_mfma_scale_f32_16x16x128_f8f6f4 : ROCDL_Mfma_OO_IntrOp<"mfma.scale.f32.16x16x128.f8f6f4", [0,1]>;
 def ROCDL_mfma_scale_f32_32x32x64_f8f6f4 : ROCDL_Mfma_OO_IntrOp<"mfma.scale.f32.32x32x64.f8f6f4", [0,1]>;
+
+// 2:4 Sparsity ops (GFX940)
+def ROCDL_smfmac_f32_16x16x32_f16 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x32.f16">;
+def ROCDL_smfmac_f32_32x32x16_f16 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x16.f16">;
+def ROCDL_smfmac_f32_16x16x32_bf16 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x32.bf16">;
+def ROCDL_smfmac_f32_32x32x16_bf16 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x23x16.bf16">;
+def ROCDL_smfmac_i32_16x16x64_i8 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x64.i8">;
+def ROCDL_smfmac_i32_32x32x32_i8 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x32.i8">;
+def ROCDL_smfmac_f32_16x16x64_bf8_bf8 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x64.bf8.bf8">;
+def ROCDL_smfmac_f32_16x16x64_bf8_fp8 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x64.bf8.fp8">;
+def ROCDL_smfmac_f32_16x16x64_fp8_bf8 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x64.fp8.fp8">;
+def ROCDL_smfmac_f32_16x16x64_fp8_fp8 : ROCDL_Mfma_IntrOp<"smfmac.f32.16x16x64.fp8.fp8">;
+def ROCDL_smfmac_f32_32x32x32_bf8_bf8 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x32.bf8.bf8">;
+def ROCDL_smfmac_f32_32x32x32_bf8_fp8 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x32.bf8.fp8">;
+def ROCDL_smfmac_f32_32x32x32_fp8_bf8 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x32.fp8.fb8">;
+def ROCDL_smfmac_f32_32x32x32_fp8_fp8 : ROCDL_Mfma_IntrOp<"smfmac.f32.32x32x32.fp8.fp8">;
+
+
 //===---------------------------------------------------------------------===//
 // WMMA intrinsics
 class ROCDL_Wmma_IntrOp<string mnemonic, list<int> overloadedOperands,