AMDGPU: Replace disabled exp inputs with undef

[arsenm]

ping

[mareko]

lib/Transforms/InstCombine/InstCombineCalls.cpp

3223

If IsCompr == true, EnBits contains 0x3 if the 1st channel is exported, 0xc if the second channel is exported, and 0xf if the first 2 channels are exported.

[arsenm]

Check full mask

[mareko]

LGTM, but maybe there should be parentheses around "2 * I".

[arsenm]

r295914

lib/Transforms/InstCombine/InstCombineCalls.cpp

if (const ConstantFP *C0 = dyn_cast<ConstantFP>(Src0)) {
           = II->getType()->getScalarType()->getFltSemantics();
         bool LosesInfo;
         APFloat Val0 = C0->getValueAPF();
         APFloat Val1 = C1->getValueAPF();
         Val0.convert(HalfSem, APFloat::rmTowardZero, &LosesInfo);
         Val1.convert(HalfSem, APFloat::rmTowardZero, &LosesInfo);
 
         Constant *Folded = ConstantVector::get({
             ConstantFP::get(II->getContext(), Val0),
             ConstantFP::get(II->getContext(), Val1) });
         return replaceInstUsesWith(*II, Folded);
       }
     }
 
     if (isa<UndefValue>(Src0) && isa<UndefValue>(Src1))
       return replaceInstUsesWith(*II, UndefValue::get(II->getType()));
 
     break;
   }
+  case Intrinsic::amdgcn_exp:
+  case Intrinsic::amdgcn_exp_compr: {
+    ConstantInt *En = dyn_cast<ConstantInt>(II->getArgOperand(1));
+    if (!En) // Illegal.
+      break;
+
+    unsigned EnBits = En->getZExtValue();
+    if (EnBits == 0xf)
+      break; // All inputs enabled.
+
+    bool IsCompr = II->getIntrinsicID() == Intrinsic::amdgcn_exp_compr;
+    bool Changed = false;
+    for (int I = 0; I < (IsCompr ? 2 : 4); ++I) {
+      if ((!IsCompr && (EnBits & (1 << I)) == 0) ||
+          (IsCompr && ((EnBits & (0x3 << 2 * I)) == 0))) {
+        Value *Src = II->getArgOperand(I + 2);
+        if (!isa<UndefValue>(Src)) {
+          II->setArgOperand(I + 2, UndefValue::get(Src->getType()));
+          Changed = true;
+        }
+      }
+    }
+
+    if (Changed)
+      return II;
+
+    break;
+  }
   case Intrinsic::stackrestore: {
     // If the save is right next to the restore, remove the restore.  This can
     // happen when variable allocas are DCE'd.
     if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
       if (SS->getIntrinsicID() == Intrinsic::stacksave) {
         if (&*++SS->getIterator() == II)
           return eraseInstFromFunction(CI);
       }

test/Transforms/InstCombine/amdgcn-intrinsics.ll

 
 ; Test constant values where rtz changes result
 ; CHECK-LABEL: @constant_rtz_pkrtz(
 ; CHECK: ret <2 x half> <half 0xH7BFF, half 0xH7BFF>
 define <2 x half> @constant_rtz_pkrtz() {
   %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 65535.0, float 65535.0)
   ret <2 x half> %cvt
 }
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.exp
+; --------------------------------------------------------------------
+
+declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) nounwind inaccessiblememonly
+
+; Make sure no crashing on invalid variable params
+; CHECK-LABEL: @exp_invalid_inputs(
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 %en, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 %tgt, i32 15, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 true, i1 false)
+define void @exp_invalid_inputs(i32 %tgt, i32 %en) {
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 %en, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 %tgt, i32 15, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  ret void
+}
+
+; CHECK-LABEL: @exp_disabled_inputs_to_undef(
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.000000e+00, float undef, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float 4.000000e+00, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float %x, float undef, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float %y, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float %z, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float %w, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float undef, float undef, float undef, float undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.000000e+00, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.000000e+00, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.000000e+00, float undef, float undef, float 4.000000e+00, i1 false, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 false, i1 false)
+define void @exp_disabled_inputs_to_undef(float %x, float %y, float %z, float %w) {
+  ; enable src0..src3 constants
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+
+  ; enable src0..src3 variables
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float %x, float %y, float %z, float %w, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float %x, float %y, float %z, float %w, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float %x, float %y, float %z, float %w, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float %x, float %y, float %z, float %w, i1 true, i1 false)
+
+  ; enable none
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float %x, float %y, float %z, float %w, i1 true, i1 false)
+
+  ; enable different source combinations
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.0, float 2.0, float 0.5, float 4.0, i1 false, i1 false)
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float 1.0, float 2.0, float 0.5, float 4.0, i1 false, i1 false)
+
+  ret void
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.exp.compr
+; --------------------------------------------------------------------
+
+declare void @llvm.amdgcn.exp.compr.v2f16(i32, i32, <2 x half>, <2 x half>, i1, i1) nounwind inaccessiblememonly
+
+; CHECK-LABEL: @exp_compr_invalid_inputs(
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 %en, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> <half 0xH3800, half 0xH4400>, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 %tgt, i32 5, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> <half 0xH3800, half 0xH4400>, i1 true, i1 false)
+define void @exp_compr_invalid_inputs(i32 %tgt, i32 %en) {
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 %en, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 %tgt, i32 5, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+  ret void
+}
+
+; CHECK-LABEL: @exp_compr_disabled_inputs_to_undef(
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> undef, <2 x half> %zw, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+define void @exp_compr_disabled_inputs_to_undef(<2 x half> %xy, <2 x half> %zw) {
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+  ret void
+}