[llvm] f7480bc - [AMDGPU] Decouple V_PK_MOV_B32 from FeaturePackedFP32Ops

[Stanislav Mekhanoshin via llvm-commits]

Author: Stanislav Mekhanoshin
Date: 2023-08-14T14:22:52-07:00
New Revision: f7480bc5c16e024291fb0036661c909d3c5d19c8

URL: https://github.com/llvm/llvm-project/commit/f7480bc5c16e024291fb0036661c909d3c5d19c8
DIFF: https://github.com/llvm/llvm-project/commit/f7480bc5c16e024291fb0036661c909d3c5d19c8.diff

LOG: [AMDGPU] Decouple V_PK_MOV_B32 from FeaturePackedFP32Ops

This is not an FP32 operation.

Differential Revision: https://reviews.llvm.org/D157909

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPU.td
    llvm/lib/Target/AMDGPU/GCNSubtarget.h
    llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
    llvm/lib/Target/AMDGPU/VOP3PInstructions.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index facaf8fd7145b3..dd30fcac2ae585 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1705,6 +1705,9 @@ def Has64BitDPP : Predicate<"Subtarget->has64BitDPP()">,
 def HasPackedFP32Ops : Predicate<"Subtarget->hasPackedFP32Ops()">,
   AssemblerPredicate<(all_of FeaturePackedFP32Ops)>;
 
+def HasPkMovB32 : Predicate<"Subtarget->hasPkMovB32()">,
+  AssemblerPredicate<(all_of FeatureGFX90AInsts)>;
+
 def HasFmaakFmamkF32Insts :
   Predicate<"Subtarget->hasFmaakFmamkF32Insts()">,
   AssemblerPredicate<(any_of FeatureGFX10Insts, FeatureGFX940Insts)>;

diff  --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index 27438e9a9fe358..55966ab5a6a160 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -916,6 +916,11 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return HasPackedFP32Ops;
   }
 
+  // Has V_PK_MOV_B32 opcode
+  bool hasPkMovB32() const {
+    return GFX90AInsts;
+  }
+
   bool hasFmaakFmamkF32Insts() const {
     return getGeneration() >= GFX10 || hasGFX940Insts();
   }

diff  --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index b0e033a1f3875e..7a35139da8cbd8 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -939,7 +939,7 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
         .addReg(SrcReg, getKillRegState(KillSrc));
       return;
     }
-    if (ST.hasPackedFP32Ops()) {
+    if (ST.hasPkMovB32()) {
       BuildMI(MBB, MI, DL, get(AMDGPU::V_PK_MOV_B32), DestReg)
         .addImm(SISrcMods::OP_SEL_1)
         .addReg(SrcReg)
@@ -986,7 +986,7 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     if (ST.hasMovB64()) {
       Opcode = AMDGPU::V_MOV_B64_e32;
       EltSize = 8;
-    } else if (ST.hasPackedFP32Ops()) {
+    } else if (ST.hasPkMovB32()) {
       Opcode = AMDGPU::V_PK_MOV_B32;
       EltSize = 8;
     }
@@ -2026,7 +2026,7 @@ bool SIInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
       APInt Imm(64, SrcOp.getImm());
       APInt Lo(32, Imm.getLoBits(32).getZExtValue());
       APInt Hi(32, Imm.getHiBits(32).getZExtValue());
-      if (ST.hasPackedFP32Ops() && Lo == Hi && isInlineConstant(Lo)) {
+      if (ST.hasPkMovB32() && Lo == Hi && isInlineConstant(Lo)) {
         BuildMI(MBB, MI, DL, get(AMDGPU::V_PK_MOV_B32), Dst)
           .addImm(SISrcMods::OP_SEL_1)
           .addImm(Lo.getSExtValue())
@@ -2047,7 +2047,7 @@ bool SIInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
       }
     } else {
       assert(SrcOp.isReg());
-      if (ST.hasPackedFP32Ops() &&
+      if (ST.hasPkMovB32() &&
           !RI.isAGPR(MBB.getParent()->getRegInfo(), SrcOp.getReg())) {
         BuildMI(MBB, MI, DL, get(AMDGPU::V_PK_MOV_B32), Dst)
           .addImm(SISrcMods::OP_SEL_1) // src0_mod

diff  --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index 71e09611e74e22..24bd2addc6652e 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -738,11 +738,15 @@ def MAIInstInfoTable : GenericTable {
 }
 
 let SubtargetPredicate = HasPackedFP32Ops, isCommutable = 1, isReMaterializable = 1 in {
-  defm V_PK_FMA_F32 : VOP3PInst<"v_pk_fma_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fma>;
-  defm V_PK_MUL_F32 : VOP3PInst<"v_pk_mul_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fmul>;
-  defm V_PK_ADD_F32 : VOP3PInst<"v_pk_add_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fadd>;
+  let SubtargetPredicate = HasPackedFP32Ops in {
+    defm V_PK_FMA_F32 : VOP3PInst<"v_pk_fma_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fma>;
+    defm V_PK_MUL_F32 : VOP3PInst<"v_pk_mul_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fmul>;
+    defm V_PK_ADD_F32 : VOP3PInst<"v_pk_add_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fadd>;
+  } // End SubtargetPredicate = HasPackedFP32Ops
+
+  let SubtargetPredicate = HasPkMovB32 in
   defm V_PK_MOV_B32 : VOP3PInst<"v_pk_mov_b32", VOP3P_Profile<VOP_V2I32_V2I32_V2I32, VOP3_PACKED>>;
-} // End SubtargetPredicate = HasPackedFP32Ops, isCommutable = 1
+} // End isCommutable = 1, isReMaterializable = 1
 
 def : MnemonicAlias<"v_accvgpr_read",  "v_accvgpr_read_b32">;
 def : MnemonicAlias<"v_accvgpr_write", "v_accvgpr_write_b32">;