Merging r195514:

------------------------------------------------------------------------
r195514 | tstellar | 2013-11-22 15:07:58 -0800 (Fri, 22 Nov 2013) | 6 lines

R600/SI: Fixing handling of condition codes

We were ignoring the ordered/onordered bits and also the signed/unsigned
bits of condition codes when lowering the DAG to MachineInstrs.

NOTE: This is a candidate for the 3.4 branch.
------------------------------------------------------------------------


git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_34@195609 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 98 insertions, 76 deletions

diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
index 3e1fc27e12..3c5375d84e 100644
--- a/lib/Target/R600/AMDGPUInstructions.td
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -37,66 +37,73 @@ class AMDGPUShaderInst <dag outs, dag ins, string asm, list<dag> pattern>
 def InstFlag : OperandWithDefaultOps <i32, (ops (i32 0))>;
 def ADDRIndirect : ComplexPattern<iPTR, 2, "SelectADDRIndirect", [], []>;
 
-def COND_EQ : PatLeaf <
-  (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETOEQ: case ISD::SETUEQ:
-                     case ISD::SETEQ: return true;}}}]
->;
+//===----------------------------------------------------------------------===//
+// PatLeafs for floating-point comparisons
+//===----------------------------------------------------------------------===//
 
 def COND_OEQ : PatLeaf <
   (cond),
   [{return N->get() == ISD::SETOEQ || N->get() == ISD::SETEQ;}]
 >;
 
-def COND_NE : PatLeaf <
+def COND_OGT : PatLeaf <
   (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETONE: case ISD::SETUNE:
-                     case ISD::SETNE: return true;}}}]
+  [{return N->get() == ISD::SETOGT || N->get() == ISD::SETGT;}]
 >;
 
-def COND_UNE : PatLeaf <
+def COND_OGE : PatLeaf <
   (cond),
-  [{return N->get() == ISD::SETUNE || N->get() == ISD::SETNE;}]
+  [{return N->get() == ISD::SETOGE || N->get() == ISD::SETGE;}]
 >;
 
-def COND_GT : PatLeaf <
+def COND_OLT : PatLeaf <
   (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETOGT: case ISD::SETUGT:
-                     case ISD::SETGT: return true;}}}]
+  [{return N->get() == ISD::SETOLT || N->get() == ISD::SETLT;}]
 >;
 
-def COND_OGT : PatLeaf <
+def COND_OLE : PatLeaf <
   (cond),
-  [{return N->get() == ISD::SETOGT || N->get() == ISD::SETGT;}]
+  [{return N->get() == ISD::SETOLE || N->get() == ISD::SETLE;}]
 >;
 
-def COND_GE : PatLeaf <
+def COND_UNE : PatLeaf <
   (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETOGE: case ISD::SETUGE:
-                     case ISD::SETGE: return true;}}}]
+  [{return N->get() == ISD::SETUNE || N->get() == ISD::SETNE;}]
 >;
 
-def COND_OGE : PatLeaf <
-  (cond),
-  [{return N->get() == ISD::SETOGE || N->get() == ISD::SETGE;}]
->;
+def COND_O : PatLeaf <(cond), [{return N->get() == ISD::SETO;}]>;
+def COND_UO : PatLeaf <(cond), [{return N->get() == ISD::SETUO;}]>;
+
+//===----------------------------------------------------------------------===//
+// PatLeafs for unsigned comparisons
+//===----------------------------------------------------------------------===//
+
+def COND_UGT : PatLeaf <(cond), [{return N->get() == ISD::SETUGT;}]>;
+def COND_UGE : PatLeaf <(cond), [{return N->get() == ISD::SETUGE;}]>;
+def COND_ULT : PatLeaf <(cond), [{return N->get() == ISD::SETULT;}]>;
+def COND_ULE : PatLeaf <(cond), [{return N->get() == ISD::SETULE;}]>;
+
+//===----------------------------------------------------------------------===//
+// PatLeafs for signed comparisons
+//===----------------------------------------------------------------------===//
+
+def COND_SGT : PatLeaf <(cond), [{return N->get() == ISD::SETGT;}]>;
+def COND_SGE : PatLeaf <(cond), [{return N->get() == ISD::SETGE;}]>;
+def COND_SLT : PatLeaf <(cond), [{return N->get() == ISD::SETLT;}]>;
+def COND_SLE : PatLeaf <(cond), [{return N->get() == ISD::SETLE;}]>;
+
+//===----------------------------------------------------------------------===//
+// PatLeafs for integer equality
+//===----------------------------------------------------------------------===//
 
-def COND_LT : PatLeaf <
+def COND_EQ : PatLeaf <
   (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETOLT: case ISD::SETULT:
-                     case ISD::SETLT: return true;}}}]
+  [{return N->get() == ISD::SETEQ || N->get() == ISD::SETUEQ;}]
 >;
 
-def COND_LE : PatLeaf <
+def COND_NE : PatLeaf <
   (cond),
-  [{switch(N->get()){{default: return false;
-                     case ISD::SETOLE: case ISD::SETULE:
-                     case ISD::SETLE: return true;}}}]
+  [{return N->get() == ISD::SETNE || N->get() == ISD::SETUNE;}]
 >;
 
 def COND_NULL : PatLeaf <
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index d8e431a8cd..abfde50173 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -822,12 +822,12 @@ def CNDE_INT : R600_3OP <
 
 def CNDGE_INT : R600_3OP <
   0x1E, "CNDGE_INT",
-  [(set i32:$dst, (selectcc i32:$src0, 0, i32:$src1, i32:$src2, COND_GE))]
+  [(set i32:$dst, (selectcc i32:$src0, 0, i32:$src1, i32:$src2, COND_SGE))]
 >;
 
 def CNDGT_INT : R600_3OP <
   0x1D, "CNDGT_INT",
-  [(set i32:$dst, (selectcc i32:$src0, 0, i32:$src1, i32:$src2, COND_GT))]
+  [(set i32:$dst, (selectcc i32:$src0, 0, i32:$src1, i32:$src2, COND_SGT))]
 >;
 
 //===----------------------------------------------------------------------===//
@@ -2313,7 +2313,7 @@ def : CND_INT_f32 <CNDGE_INT, SETGE>;
 
 //CNDGE_INT extra pattern
 def : Pat <
-  (selectcc i32:$src0, -1, i32:$src1, i32:$src2, COND_GT),
+  (selectcc i32:$src0, -1, i32:$src1, i32:$src2, COND_SGT),
   (CNDGE_INT $src0, $src1, $src2)
 >;
 
diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index 9435e9b4fc..d5d2b68caf 100644
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -56,6 +56,21 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
 
   computeRegisterProperties();
 
+  // Condition Codes
+  setCondCodeAction(ISD::SETONE, MVT::f32, Expand);
+  setCondCodeAction(ISD::SETUEQ, MVT::f32, Expand);
+  setCondCodeAction(ISD::SETUGE, MVT::f32, Expand);
+  setCondCodeAction(ISD::SETUGT, MVT::f32, Expand);
+  setCondCodeAction(ISD::SETULE, MVT::f32, Expand);
+  setCondCodeAction(ISD::SETULT, MVT::f32, Expand);
+
+  setCondCodeAction(ISD::SETONE, MVT::f64, Expand);
+  setCondCodeAction(ISD::SETUEQ, MVT::f64, Expand);
+  setCondCodeAction(ISD::SETUGE, MVT::f64, Expand);
+  setCondCodeAction(ISD::SETUGT, MVT::f64, Expand);
+  setCondCodeAction(ISD::SETULE, MVT::f64, Expand);
+  setCondCodeAction(ISD::SETULT, MVT::f64, Expand);
+
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8i32, Expand);
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8f32, Expand);
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i32, Expand);
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 855e1a9998..7827a9ac62 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -143,19 +143,19 @@ def S_GETREG_REGRD_B32 : SOPK_32 <0x00000014, "S_GETREG_REGRD_B32", []>;
 let isCompare = 1 in {
 
 defm V_CMP_F_F32 : VOPC_32 <0x00000000, "V_CMP_F_F32">;
-defm V_CMP_LT_F32 : VOPC_32 <0x00000001, "V_CMP_LT_F32", f32, COND_LT>;
-defm V_CMP_EQ_F32 : VOPC_32 <0x00000002, "V_CMP_EQ_F32", f32, COND_EQ>;
-defm V_CMP_LE_F32 : VOPC_32 <0x00000003, "V_CMP_LE_F32", f32, COND_LE>;
-defm V_CMP_GT_F32 : VOPC_32 <0x00000004, "V_CMP_GT_F32", f32, COND_GT>;
-defm V_CMP_LG_F32 : VOPC_32 <0x00000005, "V_CMP_LG_F32", f32, COND_NE>;
-defm V_CMP_GE_F32 : VOPC_32 <0x00000006, "V_CMP_GE_F32", f32, COND_GE>;
-defm V_CMP_O_F32 : VOPC_32 <0x00000007, "V_CMP_O_F32">;
-defm V_CMP_U_F32 : VOPC_32 <0x00000008, "V_CMP_U_F32">;
+defm V_CMP_LT_F32 : VOPC_32 <0x00000001, "V_CMP_LT_F32", f32, COND_OLT>;
+defm V_CMP_EQ_F32 : VOPC_32 <0x00000002, "V_CMP_EQ_F32", f32, COND_OEQ>;
+defm V_CMP_LE_F32 : VOPC_32 <0x00000003, "V_CMP_LE_F32", f32, COND_OLE>;
+defm V_CMP_GT_F32 : VOPC_32 <0x00000004, "V_CMP_GT_F32", f32, COND_OGT>;
+defm V_CMP_LG_F32 : VOPC_32 <0x00000005, "V_CMP_LG_F32">;
+defm V_CMP_GE_F32 : VOPC_32 <0x00000006, "V_CMP_GE_F32", f32, COND_OGE>;
+defm V_CMP_O_F32 : VOPC_32 <0x00000007, "V_CMP_O_F32", f32, COND_O>;
+defm V_CMP_U_F32 : VOPC_32 <0x00000008, "V_CMP_U_F32", f32, COND_UO>;
 defm V_CMP_NGE_F32 : VOPC_32 <0x00000009, "V_CMP_NGE_F32">;
 defm V_CMP_NLG_F32 : VOPC_32 <0x0000000a, "V_CMP_NLG_F32">;
 defm V_CMP_NGT_F32 : VOPC_32 <0x0000000b, "V_CMP_NGT_F32">;
 defm V_CMP_NLE_F32 : VOPC_32 <0x0000000c, "V_CMP_NLE_F32">;
-defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_NE>;
+defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_UNE>;
 defm V_CMP_NLT_F32 : VOPC_32 <0x0000000e, "V_CMP_NLT_F32">;
 defm V_CMP_TRU_F32 : VOPC_32 <0x0000000f, "V_CMP_TRU_F32">;
 
@@ -181,19 +181,19 @@ defm V_CMPX_TRU_F32 : VOPC_32 <0x0000001f, "V_CMPX_TRU_F32">;
 } // End hasSideEffects = 1, Defs = [EXEC]
 
 defm V_CMP_F_F64 : VOPC_64 <0x00000020, "V_CMP_F_F64">;
-defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64", f64, COND_LT>;
-defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64", f64, COND_EQ>;
-defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64", f64, COND_LE>;
-defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64", f64, COND_GT>;
+defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64", f64, COND_OLT>;
+defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64", f64, COND_OEQ>;
+defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64", f64, COND_OLE>;
+defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64", f64, COND_OGT>;
 defm V_CMP_LG_F64 : VOPC_64 <0x00000025, "V_CMP_LG_F64">;
-defm V_CMP_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64", f64, COND_GE>;
-defm V_CMP_O_F64 : VOPC_64 <0x00000027, "V_CMP_O_F64">;
-defm V_CMP_U_F64 : VOPC_64 <0x00000028, "V_CMP_U_F64">;
+defm V_CMP_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64", f64, COND_OGE>;
+defm V_CMP_O_F64 : VOPC_64 <0x00000027, "V_CMP_O_F64", f64, COND_O>;
+defm V_CMP_U_F64 : VOPC_64 <0x00000028, "V_CMP_U_F64", f64, COND_UO>;
 defm V_CMP_NGE_F64 : VOPC_64 <0x00000029, "V_CMP_NGE_F64">;
 defm V_CMP_NLG_F64 : VOPC_64 <0x0000002a, "V_CMP_NLG_F64">;
 defm V_CMP_NGT_F64 : VOPC_64 <0x0000002b, "V_CMP_NGT_F64">;
 defm V_CMP_NLE_F64 : VOPC_64 <0x0000002c, "V_CMP_NLE_F64">;
-defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64", f64, COND_NE>;
+defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64", f64, COND_UNE>;
 defm V_CMP_NLT_F64 : VOPC_64 <0x0000002e, "V_CMP_NLT_F64">;
 defm V_CMP_TRU_F64 : VOPC_64 <0x0000002f, "V_CMP_TRU_F64">;
 
@@ -295,12 +295,12 @@ defm V_CMPSX_TRU_F64 : VOPC_64 <0x0000007f, "V_CMPSX_TRU_F64">;
 } // End hasSideEffects = 1, Defs = [EXEC]
 
 defm V_CMP_F_I32 : VOPC_32 <0x00000080, "V_CMP_F_I32">;
-defm V_CMP_LT_I32 : VOPC_32 <0x00000081, "V_CMP_LT_I32", i32, COND_LT>;
+defm V_CMP_LT_I32 : VOPC_32 <0x00000081, "V_CMP_LT_I32", i32, COND_SLT>;
 defm V_CMP_EQ_I32 : VOPC_32 <0x00000082, "V_CMP_EQ_I32", i32, COND_EQ>;
-defm V_CMP_LE_I32 : VOPC_32 <0x00000083, "V_CMP_LE_I32", i32, COND_LE>;
-defm V_CMP_GT_I32 : VOPC_32 <0x00000084, "V_CMP_GT_I32", i32, COND_GT>;
+defm V_CMP_LE_I32 : VOPC_32 <0x00000083, "V_CMP_LE_I32", i32, COND_SLE>;
+defm V_CMP_GT_I32 : VOPC_32 <0x00000084, "V_CMP_GT_I32", i32, COND_SGT>;
 defm V_CMP_NE_I32 : VOPC_32 <0x00000085, "V_CMP_NE_I32", i32, COND_NE>;
-defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_GE>;
+defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_SGE>;
 defm V_CMP_T_I32 : VOPC_32 <0x00000087, "V_CMP_T_I32">;
 
 let hasSideEffects = 1, Defs = [EXEC] in {
@@ -317,12 +317,12 @@ defm V_CMPX_T_I32 : VOPC_32 <0x00000097, "V_CMPX_T_I32">;
 } // End hasSideEffects = 1, Defs = [EXEC]
 
 defm V_CMP_F_I64 : VOPC_64 <0x000000a0, "V_CMP_F_I64">;
-defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64">;
-defm V_CMP_EQ_I64 : VOPC_64 <0x000000a2, "V_CMP_EQ_I64">;
-defm V_CMP_LE_I64 : VOPC_64 <0x000000a3, "V_CMP_LE_I64">;
-defm V_CMP_GT_I64 : VOPC_64 <0x000000a4, "V_CMP_GT_I64">;
-defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64">;
-defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64">;
+defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64", i64, COND_SLT>;
+defm V_CMP_EQ_I64 : VOPC_64 <0x000000a2, "V_CMP_EQ_I64", i64, COND_EQ>;
+defm V_CMP_LE_I64 : VOPC_64 <0x000000a3, "V_CMP_LE_I64", i64, COND_SLE>;
+defm V_CMP_GT_I64 : VOPC_64 <0x000000a4, "V_CMP_GT_I64", i64, COND_SGT>;
+defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64", i64, COND_NE>;
+defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64", i64, COND_SGE>;
 defm V_CMP_T_I64 : VOPC_64 <0x000000a7, "V_CMP_T_I64">;
 
 let hasSideEffects = 1, Defs = [EXEC] in {
@@ -339,12 +339,12 @@ defm V_CMPX_T_I64 : VOPC_64 <0x000000b7, "V_CMPX_T_I64">;
 } // End hasSideEffects = 1, Defs = [EXEC]
 
 defm V_CMP_F_U32 : VOPC_32 <0x000000c0, "V_CMP_F_U32">;
-defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32">;
-defm V_CMP_EQ_U32 : VOPC_32 <0x000000c2, "V_CMP_EQ_U32">;
-defm V_CMP_LE_U32 : VOPC_32 <0x000000c3, "V_CMP_LE_U32">;
-defm V_CMP_GT_U32 : VOPC_32 <0x000000c4, "V_CMP_GT_U32">;
-defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32">;
-defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32">;
+defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32", i32, COND_ULT>;
+defm V_CMP_EQ_U32 : VOPC_32 <0x000000c2, "V_CMP_EQ_U32", i32, COND_EQ>;
+defm V_CMP_LE_U32 : VOPC_32 <0x000000c3, "V_CMP_LE_U32", i32, COND_ULE>;
+defm V_CMP_GT_U32 : VOPC_32 <0x000000c4, "V_CMP_GT_U32", i32, COND_UGT>;
+defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32", i32, COND_NE>;
+defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32", i32, COND_UGE>;
 defm V_CMP_T_U32 : VOPC_32 <0x000000c7, "V_CMP_T_U32">;
 
 let hasSideEffects = 1, Defs = [EXEC] in {
@@ -361,12 +361,12 @@ defm V_CMPX_T_U32 : VOPC_32 <0x000000d7, "V_CMPX_T_U32">;
 } // End hasSideEffects = 1, Defs = [EXEC]
 
 defm V_CMP_F_U64 : VOPC_64 <0x000000e0, "V_CMP_F_U64">;
-defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64">;
-defm V_CMP_EQ_U64 : VOPC_64 <0x000000e2, "V_CMP_EQ_U64">;
-defm V_CMP_LE_U64 : VOPC_64 <0x000000e3, "V_CMP_LE_U64">;
-defm V_CMP_GT_U64 : VOPC_64 <0x000000e4, "V_CMP_GT_U64">;
-defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64">;
-defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64">;
+defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64", i64, COND_ULT>;
+defm V_CMP_EQ_U64 : VOPC_64 <0x000000e2, "V_CMP_EQ_U64", i64, COND_EQ>;
+defm V_CMP_LE_U64 : VOPC_64 <0x000000e3, "V_CMP_LE_U64", i64, COND_ULE>;
+defm V_CMP_GT_U64 : VOPC_64 <0x000000e4, "V_CMP_GT_U64", i64, COND_UGT>;
+defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64", i64, COND_NE>;
+defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64", i64, COND_UGE>;
 defm V_CMP_T_U64 : VOPC_64 <0x000000e7, "V_CMP_T_U64">;
 
 let hasSideEffects = 1, Defs = [EXEC] in {