[llvm-branch-commits] [llvm] [AMDGPU] Add wave reduce intrinsics for double types - 2 (PR #170812)

[Juan Manuel Martinez Caamaño via llvm-branch-commits]

================
@@ -6093,12 +6150,16 @@ SITargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
     return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::S_ADD_U64_PSEUDO);
   case AMDGPU::WAVE_REDUCE_FADD_PSEUDO_F32:
     return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::V_ADD_F32_e64);
+  case AMDGPU::WAVE_REDUCE_FADD_PSEUDO_F64:
+    return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::V_ADD_F64_e64);
   case AMDGPU::WAVE_REDUCE_SUB_PSEUDO_I32:
     return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::S_SUB_I32);
   case AMDGPU::WAVE_REDUCE_SUB_PSEUDO_U64:
     return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::S_SUB_U64_PSEUDO);
   case AMDGPU::WAVE_REDUCE_FSUB_PSEUDO_F32:
     return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::V_SUB_F32_e64);
+  case AMDGPU::WAVE_REDUCE_FSUB_PSEUDO_F64:
+    return lowerWaveReduce(MI, *BB, *getSubtarget(), AMDGPU::V_SUB_F32_e32);
----------------
jmmartinez wrote:

I like `WAVE_REDUCE_FSUB_PSEUDO_F64`. It highlights that there is something different going on. When skimming through the code, `V_SUB_F32_e32` looks like an accidental typo.

https://github.com/llvm/llvm-project/pull/170812