AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 (llvm#128123)

jwanggit86 · Feb 21, 2025 · cc46d00 · cc46d00
1 parent e729dc7
commit cc46d00
Show file tree

Hide file tree

Showing 4 changed files with 112 additions and 245 deletions.
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -13515,7 +13515,8 @@ static bool supportsMin3Max3(const GCNSubtarget &Subtarget, unsigned Opc,
   case ISD::FMINIMUM:
   case ISD::FMAXIMUM:
     return (VT == MVT::f32 && Subtarget.hasMinimum3Maximum3F32()) ||
-           (VT == MVT::f16 && Subtarget.hasMinimum3Maximum3F16());
+           (VT == MVT::f16 && Subtarget.hasMinimum3Maximum3F16()) ||
+           (VT == MVT::v2f16 && Subtarget.hasMinimum3Maximum3PKF16());
   case ISD::SMAX:
   case ISD::SMIN:
   case ISD::UMAX:

diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -145,8 +145,8 @@ def : VOP3PSatPat<ssubsat, V_PK_SUB_I16>;
 } // End SubtargetPredicate = HasVOP3PInsts
 
 let SubtargetPredicate = HasMinimum3Maximum3PKF16, FPDPRounding = 1 in {
-defm V_PK_MINIMUM3_F16 : VOP3PInst<"v_pk_minimum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
-defm V_PK_MAXIMUM3_F16 : VOP3PInst<"v_pk_maximum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
+defm V_PK_MINIMUM3_F16 : VOP3PInst<"v_pk_minimum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, AMDGPUfminimum3>;
+defm V_PK_MAXIMUM3_F16 : VOP3PInst<"v_pk_maximum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, AMDGPUfmaximum3>;
 }
 
 // TODO: Make sure we're doing the right thing with denormals. Note