AMDGPU: Correct definitions for bitset instructions

These really read and write the result register, so these need a tied
input.

llvm-svn: 354809

diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
index 2b1264d6f6313d69924f528685e913c312649e43..e589f60b5838c11bd8f4e008a7823202d276f471 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
+++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
@@ -276,7 +276,9 @@ static bool shrinkScalarLogicOp(const GCNSubtarget &ST,
         if (Opc == AMDGPU::S_BITSET0_B32 ||
             Opc == AMDGPU::S_BITSET1_B32) {
           Src0->ChangeToImmediate(NewImm);
-          MI.RemoveOperand(2);
+          // Remove the immediate and add the tied input.
+          MI.getOperand(2).ChangeToRegister(Dest->getReg(), false);
+          MI.tieOperands(0, 2);
         } else {
           SrcImm->setImm(NewImm);
         }

diff --git a/llvm/lib/Target/AMDGPU/SOPInstructions.td b/llvm/lib/Target/AMDGPU/SOPInstructions.td
index 1cb9f27dd021624188da8068345829ac637f8f00..5296f24f5e4f1f2cd37a4d702b7dc0eff272285f 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -76,10 +76,13 @@ class SOP1_Real<bits<8> op, SOP1_Pseudo ps> :
   let Inst{31-23} = 0x17d; //encoding;
 }
 
-class SOP1_32 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
-  opName, (outs SReg_32:$sdst), (ins SSrc_b32:$src0),
-  "$sdst, $src0", pattern
->;
+class SOP1_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo <
+  opName, (outs SReg_32:$sdst),
+  !if(tied_in, (ins SSrc_b32:$src0, SReg_32:$sdst_in),
+               (ins SSrc_b32:$src0)),
+  "$sdst, $src0", pattern> {
+  let Constraints = !if(tied_in, "$sdst = $sdst_in", "");
+}
 
 // 32-bit input, no output.
 class SOP1_0_32 <string opName, list<dag> pattern = []> : SOP1_Pseudo <
@@ -106,10 +109,13 @@ class SOP1_32_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
 >;
 
 // 32-bit input, 64-bit output.
-class SOP1_64_32 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
-  opName, (outs SReg_64:$sdst), (ins SSrc_b32:$src0),
-  "$sdst, $src0", pattern
->;
+class SOP1_64_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo <
+  opName, (outs SReg_64:$sdst),
+  !if(tied_in, (ins SSrc_b32:$src0, SReg_64:$sdst_in),
+               (ins SSrc_b32:$src0)),
+  "$sdst, $src0", pattern> {
+  let Constraints = !if(tied_in, "$sdst = $sdst_in", "");
+}
 
 // no input, 64-bit output.
 class SOP1_64_0 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
@@ -189,10 +195,10 @@ def S_SEXT_I32_I16 : SOP1_32 <"s_sext_i32_i16",
   [(set i32:$sdst, (sext_inreg i32:$src0, i16))]
 >;
 
-def S_BITSET0_B32 : SOP1_32    <"s_bitset0_b32">;
-def S_BITSET0_B64 : SOP1_64_32 <"s_bitset0_b64">;
-def S_BITSET1_B32 : SOP1_32    <"s_bitset1_b32">;
-def S_BITSET1_B64 : SOP1_64_32 <"s_bitset1_b64">;
+def S_BITSET0_B32 : SOP1_32    <"s_bitset0_b32", [], 1>;
+def S_BITSET0_B64 : SOP1_64_32 <"s_bitset0_b64", [], 1>;
+def S_BITSET1_B32 : SOP1_32    <"s_bitset1_b32", [], 1>;
+def S_BITSET1_B64 : SOP1_64_32 <"s_bitset1_b64", [], 1>;
 def S_GETPC_B64 : SOP1_64_0  <"s_getpc_b64",
   [(set i64:$sdst, (int_amdgcn_s_getpc))]
 >;