Fixed a bug in printing "cmp" pseudo ops.

> This IR code
> %res = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a1, i8 14)
> fails with assertion:
>
> llc: X86ATTInstPrinter.cpp:62: void llvm::X86ATTInstPrinter::printSSECC(const llvm::MCInst*, unsigned int, llvm::raw_ostream&): Assertion `0 && "Invalid ssecc argument!"' failed.
> 0  llc             0x0000000001355803
> 1  llc             0x0000000001355dc9
> 2  libpthread.so.0 0x00007f79a30575d0
> 3  libc.so.6       0x00007f79a23a1945 gsignal + 53
> 4  libc.so.6       0x00007f79a23a2f21 abort + 385
> 5  libc.so.6       0x00007f79a239a810 __assert_fail + 240
> 6  llc             0x00000000011858d5 llvm::X86ATTInstPrinter::printSSECC(llvm::MCInst const*, unsigned int, llvm::raw_ostream&) + 119

I added the full testing for all possible pseudo-ops of cmp.
I extended X86AsmPrinter.cpp and X86IntelInstPrinter.cpp.

You'l also see lines alignments (unrelated to this fix) in X86IselLowering.cpp from my previous check-in.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@150068 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 183 insertions, 41 deletions

diff --git a/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp b/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
index 106f716aa5..772fe1bf19 100644
--- a/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
+++ b/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
@@ -60,14 +60,38 @@ void X86ATTInstPrinter::printSSECC(const MCInst *MI, unsigned Op,
                                    raw_ostream &O) {
   switch (MI->getOperand(Op).getImm()) {
   default: llvm_unreachable("Invalid ssecc argument!");
-  case 0: O << "eq"; break;
-  case 1: O << "lt"; break;
-  case 2: O << "le"; break;
-  case 3: O << "unord"; break;
-  case 4: O << "neq"; break;
-  case 5: O << "nlt"; break;
-  case 6: O << "nle"; break;
-  case 7: O << "ord"; break;
+  case    0: O << "eq"; break;
+  case    1: O << "lt"; break;
+  case    2: O << "le"; break;
+  case    3: O << "unord"; break;
+  case    4: O << "neq"; break;
+  case    5: O << "nlt"; break;
+  case    6: O << "nle"; break;
+  case    7: O << "ord"; break;
+  case    8: O << "eq_uq"; break;
+  case    9: O << "nge"; break;
+  case  0xa: O << "ngt"; break;
+  case  0xb: O << "false"; break;
+  case  0xc: O << "neq_oq"; break;
+  case  0xd: O << "ge"; break;
+  case  0xe: O << "gt"; break;
+  case  0xf: O << "true"; break;
+  case 0x10: O << "eq_os"; break;
+  case 0x11: O << "lt_oq"; break;
+  case 0x12: O << "le_oq"; break;
+  case 0x13: O << "unord_s"; break;
+  case 0x14: O << "neq_us"; break;
+  case 0x15: O << "nlt_uq"; break;
+  case 0x16: O << "nle_uq"; break;
+  case 0x17: O << "ord_s"; break;
+  case 0x18: O << "eq_us"; break;
+  case 0x19: O << "nge_uq"; break;
+  case 0x1a: O << "ngt_uq"; break;
+  case 0x1b: O << "false_os"; break;
+  case 0x1c: O << "neq_os"; break;
+  case 0x1d: O << "ge_oq"; break;
+  case 0x1e: O << "gt_oq"; break;
+  case 0x1f: O << "true_us"; break;
   }
 }
 
diff --git a/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp b/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
index 3bbe127338..7f1f8a3c4c 100644
--- a/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
+++ b/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
@@ -50,14 +50,39 @@ void X86IntelInstPrinter::printSSECC(const MCInst *MI, unsigned Op,
                                      raw_ostream &O) {
   switch (MI->getOperand(Op).getImm()) {
   default: llvm_unreachable("Invalid ssecc argument!");
-  case 0: O << "eq"; break;
-  case 1: O << "lt"; break;
-  case 2: O << "le"; break;
-  case 3: O << "unord"; break;
-  case 4: O << "neq"; break;
-  case 5: O << "nlt"; break;
-  case 6: O << "nle"; break;
-  case 7: O << "ord"; break;
+  case    0: O << "eq"; break;
+  case    1: O << "lt"; break;
+  case    2: O << "le"; break;
+  case    3: O << "unord"; break;
+  case    4: O << "neq"; break;
+  case    5: O << "nlt"; break;
+  case    6: O << "nle"; break;
+  case    7: O << "ord"; break;
+  case    8: O << "eq_uq"; break;
+  case    9: O << "nge"; break;
+  case  0xa: O << "ngt"; break;
+  case  0xb: O << "false"; break;
+  case  0xc: O << "neq_oq"; break;
+  case  0xd: O << "ge"; break;
+  case  0xe: O << "gt"; break;
+  case  0xf: O << "true"; break;
+  case 0x10: O << "eq_os"; break;
+  case 0x11: O << "lt_oq"; break;
+  case 0x12: O << "le_oq"; break;
+  case 0x13: O << "unord_s"; break;
+  case 0x14: O << "neq_us"; break;
+  case 0x15: O << "nlt_uq"; break;
+  case 0x16: O << "nle_uq"; break;
+  case 0x17: O << "ord_s"; break;
+  case 0x18: O << "eq_us"; break;
+  case 0x19: O << "nge_uq"; break;
+  case 0x1a: O << "ngt_uq"; break;
+  case 0x1b: O << "false_os"; break;
+  case 0x1c: O << "neq_os"; break;
+  case 0x1d: O << "ge_oq"; break;
+  case 0x1e: O << "gt_oq"; break;
+  case 0x1f: O << "true_us"; break;
+
   }
 }
 
diff --git a/lib/Target/X86/X86AsmPrinter.cpp b/lib/Target/X86/X86AsmPrinter.cpp
index 4c3ff02826..5949064ee3 100644
--- a/lib/Target/X86/X86AsmPrinter.cpp
+++ b/lib/Target/X86/X86AsmPrinter.cpp
@@ -266,14 +266,38 @@ void X86AsmPrinter::printSSECC(const MachineInstr *MI, unsigned Op,
   unsigned char value = MI->getOperand(Op).getImm();
   assert(value <= 7 && "Invalid ssecc argument!");
   switch (value) {
-  case 0: O << "eq"; break;
-  case 1: O << "lt"; break;
-  case 2: O << "le"; break;
-  case 3: O << "unord"; break;
-  case 4: O << "neq"; break;
-  case 5: O << "nlt"; break;
-  case 6: O << "nle"; break;
-  case 7: O << "ord"; break;
+  case    0: O << "eq"; break;
+  case    1: O << "lt"; break;
+  case    2: O << "le"; break;
+  case    3: O << "unord"; break;
+  case    4: O << "neq"; break;
+  case    5: O << "nlt"; break;
+  case    6: O << "nle"; break;
+  case    7: O << "ord"; break;
+  case    8: O << "eq_uq"; break;
+  case    9: O << "nge"; break;
+  case  0xa: O << "ngt"; break;
+  case  0xb: O << "false"; break;
+  case  0xc: O << "neq_oq"; break;
+  case  0xd: O << "ge"; break;
+  case  0xe: O << "gt"; break;
+  case  0xf: O << "true"; break;
+  case 0x10: O << "eq_os"; break;
+  case 0x11: O << "lt_oq"; break;
+  case 0x12: O << "le_oq"; break;
+  case 0x13: O << "unord_s"; break;
+  case 0x14: O << "neq_us"; break;
+  case 0x15: O << "nlt_uq"; break;
+  case 0x16: O << "nle_uq"; break;
+  case 0x17: O << "ord_s"; break;
+  case 0x18: O << "eq_us"; break;
+  case 0x19: O << "nge_uq"; break;
+  case 0x1a: O << "ngt_uq"; break;
+  case 0x1b: O << "false_os"; break;
+  case 0x1c: O << "neq_os"; break;
+  case 0x1d: O << "ge_oq"; break;
+  case 0x1e: O << "gt_oq"; break;
+  case 0x1f: O << "true_us"; break;
   }
 }
 
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 7660b95369..5f4e4183d8 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -14597,41 +14597,42 @@ static SDValue PerformSExtCombine(SDNode *N, SelectionDAG &DAG,
   if (!DCI.isBeforeLegalizeOps())
     return SDValue();
 
-  if (!Subtarget->hasAVX()) return SDValue();
+  if (!Subtarget->hasAVX()) 
+    return SDValue();
 
-   // Optimize vectors in AVX mode
-   // Sign extend  v8i16 to v8i32 and
-   //              v4i32 to v4i64
-   //
-   // Divide input vector into two parts
-   // for v4i32 the shuffle mask will be { 0, 1, -1, -1} {2, 3, -1, -1}
-   // use vpmovsx instruction to extend v4i32 -> v2i64; v8i16 -> v4i32
-   // concat the vectors to original VT
+  // Optimize vectors in AVX mode
+  // Sign extend  v8i16 to v8i32 and
+  //              v4i32 to v4i64
+  //
+  // Divide input vector into two parts
+  // for v4i32 the shuffle mask will be { 0, 1, -1, -1} {2, 3, -1, -1}
+  // use vpmovsx instruction to extend v4i32 -> v2i64; v8i16 -> v4i32
+  // concat the vectors to original VT
 
   EVT VT = N->getValueType(0);
   SDValue Op = N->getOperand(0);
   EVT OpVT = Op.getValueType();
   DebugLoc dl = N->getDebugLoc();
 
-  if (((VT == MVT::v4i64) && (OpVT == MVT::v4i32)) ||
-    ((VT == MVT::v8i32) && (OpVT == MVT::v8i16))) {
+  if ((VT == MVT::v4i64 && OpVT == MVT::v4i32) ||
+      (VT == MVT::v8i32 && OpVT == MVT::v8i16)) {
 
     unsigned NumElems = OpVT.getVectorNumElements();
     SmallVector<int,8> ShufMask1(NumElems, -1);
-    for (unsigned i=0; i< NumElems/2; i++) ShufMask1[i] = i;
+    for (unsigned i = 0; i < NumElems/2; i++) ShufMask1[i] = i;
 
     SDValue OpLo = DAG.getVectorShuffle(OpVT, dl, Op, DAG.getUNDEF(OpVT),
-                                ShufMask1.data());
+                                        ShufMask1.data());
 
     SmallVector<int,8> ShufMask2(NumElems, -1);
-    for (unsigned i=0; i< NumElems/2; i++) ShufMask2[i] = i+NumElems/2;
+    for (unsigned i = 0; i < NumElems/2; i++) ShufMask2[i] = i + NumElems/2;
 
     SDValue OpHi = DAG.getVectorShuffle(OpVT, dl, Op, DAG.getUNDEF(OpVT),
-                                ShufMask2.data());
+                                        ShufMask2.data());
 
     EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), VT.getScalarType(), 
-      VT.getVectorNumElements()/2);
-    
+                                  VT.getVectorNumElements()/2);
+
     OpLo = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpLo); 
     OpHi = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpHi);
 
diff --git a/test/CodeGen/X86/avx-intrinsics-x86.ll b/test/CodeGen/X86/avx-intrinsics-x86.ll
index c02cae11c6..616601a943 100644
--- a/test/CodeGen/X86/avx-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx-intrinsics-x86.ll
@@ -1766,6 +1766,74 @@ define <8 x float> @test_x86_avx_cmp_ps_256(<8 x float> %a0, <8 x float> %a1) {
   %res = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a1, i8 7) ; <<8 x float>> [#uses=1]
   ret <8 x float> %res
 }
+
+define <8 x float> @test_x86_avx_cmp_ps_256_pseudo_op(<8 x float> %a0, <8 x float> %a1) {
+  ; CHECK: vcmpeqps
+  %a2 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a1, i8 0) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpltps
+  %a3 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a2, i8 1) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpleps
+  %a4 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a3, i8 2) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpunordps
+  %a5 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a4, i8 3) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpneqps
+  %a6 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a5, i8 4) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpnltps
+  %a7 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a6, i8 5) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpnleps
+  %a8 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a7, i8 6) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpordps
+  %a9 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a8, i8 7) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpeq_uqps
+  %a10 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a9, i8 8) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpngeps
+  %a11 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a10, i8 9) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpngtps
+  %a12 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a11, i8 10) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpfalseps
+  %a13 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a12, i8 11) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpneq_oqps
+  %a14 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a13, i8 12) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpgeps
+  %a15 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a14, i8 13) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpgtps
+  %a16 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a15, i8 14) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmptrueps
+  %a17 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a16, i8 15) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpeq_osps
+  %a18 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a17, i8 16) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmplt_oqps
+  %a19 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a18, i8 17) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmple_oqps
+  %a20 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a19, i8 18) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpunord_sps
+  %a21 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a20, i8 19) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpneq_usps
+  %a22 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a21, i8 20) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpnlt_uqps
+  %a23 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a22, i8 21) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpnle_uqps
+  %a24 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a23, i8 22) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpord_sps
+  %a25 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a24, i8 23) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpeq_usps
+  %a26 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a25, i8 24) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpnge_uqps
+  %a27 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a26, i8 25) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpngt_uqps
+  %a28 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a27, i8 26) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpfalse_osps
+  %a29 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a28, i8 27) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpneq_osps
+  %a30 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a29, i8 28) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpge_oqps
+  %a31 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a30, i8 29) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmpgt_oqps
+  %a32 = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a31, i8 30) ; <<8 x float>> [#uses=1]
+  ; CHECK: vcmptrue_usps
+  %res = call <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float> %a0, <8 x float> %a32, i8 31) ; <<8 x float>> [#uses=1]
+  ret <8 x float> %res
+}
 declare <8 x float> @llvm.x86.avx.cmp.ps.256(<8 x float>, <8 x float>, i8) nounwind readnone