[llvm-branch-commits] [llvm] AMDGPU: Codegen for v_dual_dot2acc_f32_f16/bf16 from VOP3 (PR #179226)

[Mirko Brkušanin via llvm-branch-commits]

================
@@ -2065,6 +2065,14 @@ bool SIInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
   const AMDGPU::LaneMaskConstants &LMC = AMDGPU::LaneMaskConstants::get(ST);
   switch (MI.getOpcode()) {
   default: return TargetInstrInfo::expandPostRAPseudo(MI);
+  case AMDGPU::V_DOT2ACC_F32_F16_PSEUDO:
+    MI.setDesc(get(AMDGPU::V_DOT2_F32_F16));
+    break;
+
+  case AMDGPU::V_DOT2ACC_F32_BF16_PSEUDO:
+    MI.setDesc(get(AMDGPU::V_DOT2_F32_BF16));
+    break;
----------------
mbrkusanin wrote:

I looked at print-after-all
src2 should also be untied from dst: `MI.untieRegOperand(Idx)`

https://github.com/llvm/llvm-project/pull/179226