[llvm] [AMDGPU] Add MovB64 subtarget feature (PR #196340)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Mariusz Sikora (mariusz-sikora-at-amd)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/196340.diff


4 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPU.td (+6) 
- (modified) llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp (+1-1) 
- (modified) llvm/lib/Target/AMDGPU/GCNSubtarget.h (-2) 
- (modified) llvm/lib/Target/AMDGPU/SIInstrInfo.cpp (+4-4) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index 25fc64d178858..35d31571b4592 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -483,6 +483,10 @@ defm SBarrierLeaveImm : AMDGPUSubtargetFeature<"s-barrier-leave-imm",
   "s_barrier_leave takes an immediate operand"
 >;
 
+defm MovB64Inst : AMDGPUSubtargetFeature<"mov-b64-inst",
+  "Has mov_b64 instruction"
+>;
+
 defm GFX950Insts : AMDGPUSubtargetFeature<"gfx950-insts",
   "Additional instructions for GFX950+",
   /*GenPredicate=*/1,
@@ -1755,6 +1759,7 @@ def FeatureISAVersion9_4_Common : FeatureSet<
    FeatureMemoryAtomicFAddF32DenormalSupport,
    FeatureFlatBufferGlobalAtomicFaddF64Inst,
    FeatureLshlAddU64Inst,
+   FeatureMovB64Inst,
    ]>;
 
 def FeatureISAVersion9_5_Common : FeatureSet<
@@ -2110,6 +2115,7 @@ def FeatureISAVersion12_50_Common : FeatureSet<
    FeatureMcastLoadInsts,
    FeatureNoF16PseudoScalarTransInlineConstants,
    FeatureRealTrue16Insts,
+   FeatureMovB64Inst,
 ]>;
 
 def FeatureISAVersion12_50 : FeatureSet<
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index e05213b2aaf93..37ff096a0c294 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -5551,7 +5551,7 @@ bool AMDGPUTargetLowering::isInt64ImmLegal(SDNode *N, SelectionDAG &DAG) const {
   auto &ST = DAG.getSubtarget<GCNSubtarget>();
   const auto *TII = ST.getInstrInfo();
 
-  if (!ST.hasMovB64() || (!SDConstant && !SDFPConstant))
+  if (!ST.hasMovB64Inst() || (!SDConstant && !SDFPConstant))
     return false;
 
   if (ST.has64BitLiterals())
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index ec7dd92d6b10e..2e2797269fbcf 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -507,8 +507,6 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
 
   bool hasMadF16() const;
 
-  bool hasMovB64() const { return HasGFX940Insts || HasGFX1250Insts; }
-
   // Scalar and global loads support scale_offset bit.
   bool hasScaleOffset() const { return HasGFX1250Insts; }
 
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 65137fa09e209..0dcebd43f9866 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -1041,7 +1041,7 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
   }
 
   if (RC == RI.getVGPR64Class() && (SrcRC == RC || RI.isSGPRClass(SrcRC))) {
-    if (ST.hasMovB64()) {
+    if (ST.hasMovB64Inst()) {
       BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B64_e32), DestReg)
         .addReg(SrcReg, getKillRegState(KillSrc));
       return;
@@ -1090,7 +1090,7 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
              (RI.isProperlyAlignedRC(*RC) &&
               (SrcRC == RC || RI.isSGPRClass(SrcRC)))) {
     // TODO: In 96-bit case, could do a 64-bit mov and then a 32-bit mov.
-    if (ST.hasMovB64()) {
+    if (ST.hasMovB64Inst()) {
       Opcode = AMDGPU::V_MOV_B64_e32;
       EltSize = 8;
     } else if (ST.hasPkMovB32()) {
@@ -2182,7 +2182,7 @@ bool SIInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
     const MachineOperand &SrcOp = MI.getOperand(1);
     // FIXME: Will this work for 64-bit floating point immediates?
     assert(!SrcOp.isFPImm());
-    if (ST.hasMovB64() && Mov64RC->contains(Dst)) {
+    if (ST.hasMovB64Inst() && Mov64RC->contains(Dst)) {
       MI.setDesc(Mov64Desc);
       if (SrcOp.isReg() || isInlineConstant(MI, 1) ||
           isUInt<32>(SrcOp.getImm()) || ST.has64BitLiterals())
@@ -2726,7 +2726,7 @@ std::pair<MachineInstr*, MachineInstr*>
 SIInstrInfo::expandMovDPP64(MachineInstr &MI) const {
   assert (MI.getOpcode() == AMDGPU::V_MOV_B64_DPP_PSEUDO);
 
-  if (ST.hasMovB64() && ST.hasFeature(AMDGPU::FeatureDPALU_DPP) &&
+  if (ST.hasMovB64Inst() && ST.hasFeature(AMDGPU::FeatureDPALU_DPP) &&
       AMDGPU::isLegalDPALU_DPPControl(
           ST, getNamedOperand(MI, AMDGPU::OpName::dpp_ctrl)->getImm())) {
     MI.setDesc(get(AMDGPU::V_MOV_B64_dpp));

``````````

</details>


https://github.com/llvm/llvm-project/pull/196340