[AMDGPU] Fix an issue in `op_sel_hi` skipping.

Summary:
- Only apply packed literal `op_sel_hi` skipping on operands requiring
  packed literals. Even an instruction is `packed`, it may have operand
  requiring non-packed literal, such as `v_dot2_f32_f16`.

Reviewers: rampitec, arsenm, kzhuravl

Subscribers: jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60978

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358922 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SIFoldOperands.cpp b/lib/Target/AMDGPU/SIFoldOperands.cpp
index 2ebcbae90bbf8218fc61f643c89066803b97fb0c..f5e60819f3059144d9f85c85102a4693a247deba 100644 (file)
--- a/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -189,15 +189,24 @@ static bool updateOperand(FoldCandidate &Fold,
       unsigned Val = Mod.getImm();
       if ((Val & SISrcMods::OP_SEL_0) || !(Val & SISrcMods::OP_SEL_1))
         return false;
-      // If upper part is all zero we do not need op_sel_hi.
-      if (!isUInt<16>(Fold.ImmToFold)) {
-        if (!(Fold.ImmToFold & 0xffff)) {
-          Mod.setImm(Mod.getImm() | SISrcMods::OP_SEL_0);
+      // Only apply the following transformation if that operand requries
+      // a packed immediate.
+      switch (TII.get(Opcode).OpInfo[OpNo].OperandType) {
+      case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
+      case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
+        // If upper part is all zero we do not need op_sel_hi.
+        if (!isUInt<16>(Fold.ImmToFold)) {
+          if (!(Fold.ImmToFold & 0xffff)) {
+            Mod.setImm(Mod.getImm() | SISrcMods::OP_SEL_0);
+            Mod.setImm(Mod.getImm() & ~SISrcMods::OP_SEL_1);
+            Old.ChangeToImmediate((Fold.ImmToFold >> 16) & 0xffff);
+            return true;
+          }
           Mod.setImm(Mod.getImm() & ~SISrcMods::OP_SEL_1);
-          Old.ChangeToImmediate((Fold.ImmToFold >> 16) & 0xffff);
-          return true;
         }
-        Mod.setImm(Mod.getImm() & ~SISrcMods::OP_SEL_1);
+        break;
+      default:
+        break;
       }
     }
 

diff --git a/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.ll b/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.ll
index 7efb1850a277622cf5b02acd19e5913b41de396b..e16450bfd3c20f9ea8eee9094dd2c8f9d6353a5c 100644 (file)
--- a/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.ll
+++ b/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.ll
@@ -33,3 +33,10 @@ entry:
   store float %r.val, float addrspace(1)* %r
   ret void
 }
+
+; GFX906-LABEL: {{^}}fdot2_inline_literal
+; GFX906: v_dot2_f32_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 1.0
+define float @fdot2_inline_literal(<2 x half> %a, <2 x half> %b) {
+  %ret = tail call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %b, float 1.0, i1 false)
+  ret float %ret
+}