R600/SI: fix VOP3b encoding v2

v2: document why we hardcode VCC for now.

This is a candidate for the mesa-stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 176099

diff --git a/llvm/lib/Target/R600/SIInstrInfo.td b/llvm/lib/Target/R600/SIInstrInfo.td
index 99168ce..d68fbff 100644 (file)
--- a/llvm/lib/Target/R600/SIInstrInfo.td
+++ b/llvm/lib/Target/R600/SIInstrInfo.td
@@ -51,6 +51,7 @@ class InlineImm <ValueType vt> : ImmLeaf <vt, [{
 
 def SIOperand {
   int ZERO = 0x80;
+  int VCC = 0x6A;
 }
 
 class GPR4Align <RegisterClass rc> : Operand <vAny> {
@@ -195,6 +196,29 @@ multiclass VOP2_32 <bits<6> op, string opName, list<dag> pattern>
 multiclass VOP2_64 <bits<6> op, string opName, list<dag> pattern>
   : VOP2_Helper <op, VReg_64, VSrc_64, opName, pattern>;
 
+multiclass VOP2b_32 <bits<6> op, string opName, list<dag> pattern> {
+
+  def _e32 : VOP2 <
+    op, (outs VReg_32:$dst), (ins VSrc_32:$src0, VReg_32:$src1),
+    opName#"_e32 $dst, $src0, $src1", pattern
+  >;
+
+  def _e64 : VOP3b <
+    {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+    (outs VReg_32:$dst),
+    (ins VSrc_32:$src0, VReg_32:$src1,
+         i32imm:$abs, i32imm:$clamp,
+         i32imm:$omod, i32imm:$neg),
+    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg", []
+  > {
+    let SRC2 = SIOperand.ZERO;
+    /* the VOP2 variant puts the carry out into VCC, the VOP3 variant
+       can write it into any SGPR. We currently don't use the carry out,
+       so for now hardcode it to VCC as well */
+    let SDST = SIOperand.VCC;
+  }
+}
+
 multiclass VOPC_Helper <bits<8> op, RegisterClass vrc, RegisterClass arc,
                         string opName, ValueType vt, PatLeaf cond> {
 

diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td
index 9701d19..f999025 100644 (file)
--- a/llvm/lib/Target/R600/SIInstructions.td
+++ b/llvm/lib/Target/R600/SIInstructions.td
@@ -805,17 +805,19 @@ defm V_MADAK_F32 : VOP2_32 <0x00000021, "V_MADAK_F32", []>;
 //defm V_MBCNT_LO_U32_B32 : VOP2_32 <0x00000023, "V_MBCNT_LO_U32_B32", []>;
 //defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>;
 let Defs = [VCC] in { // Carry-out goes to VCC
-defm V_ADD_I32 : VOP2_32 <0x00000025, "V_ADD_I32",
+defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32",
   [(set VReg_32:$dst, (add (i32 VSrc_32:$src0), (i32 VReg_32:$src1)))]
 >;
-defm V_SUB_I32 : VOP2_32 <0x00000026, "V_SUB_I32",
+defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32",
   [(set VReg_32:$dst, (sub (i32 VSrc_32:$src0), (i32 VReg_32:$src1)))]
 >;
+defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", []>;
+let Uses = [VCC] in { // Carry-out comes from VCC
+defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", []>;
+defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", []>;
+defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", []>;
+} // End Uses = [VCC]
 } // End Defs = [VCC]
-defm V_SUBREV_I32 : VOP2_32 <0x00000027, "V_SUBREV_I32", []>;
-defm V_ADDC_U32 : VOP2_32 <0x00000028, "V_ADDC_U32", []>;
-defm V_SUBB_U32 : VOP2_32 <0x00000029, "V_SUBB_U32", []>;
-defm V_SUBBREV_U32 : VOP2_32 <0x0000002a, "V_SUBBREV_U32", []>;
 defm V_LDEXP_F32 : VOP2_32 <0x0000002b, "V_LDEXP_F32", []>;
 ////def V_CVT_PKACCUM_U8_F32 : VOP2_U8 <0x0000002c, "V_CVT_PKACCUM_U8_F32", []>;
 ////def V_CVT_PKNORM_I16_F32 : VOP2_I16 <0x0000002d, "V_CVT_PKNORM_I16_F32", []>;