[Clang][AVX512][intrinsics] Fix rcp and sqrt intrinsics.

[m_zuckerman]

Summary

Separating between scalar intrinsics and vector intrinsics.

Bug description:
Clang lowers the rcp14_sd,rcp14_ss,rcqrt14_sd and rcqrt14_ss to a wrong instruction. This patch fixes this problem by adding new SDnode name. By that LLVM distinguishes between the two intrinsics.

lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
 
       /// Commutative FMIN and FMAX.
       FMAXC, FMINC,
 
       /// Floating point reciprocal-sqrt and reciprocal approximation.
       /// Note that these typically require refinement
       /// in order to obtain suitable precision.
       FRSQRT, FRCP,
+      FRSQRTS, FRCPS,
     
       // Thread Local Storage.
       TLSADDR,
 
       // Thread Local Storage. A call to get the start address
       // of the TLS block for the current module.
       TLSBASEADDR,
 
       // Thread Local Storage.  When calling to an OS provided

lib/Target/X86/X86ISelLowering.cpp

const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::CONFLICT:           return "X86ISD::CONFLICT";
   case X86ISD::FMAX:               return "X86ISD::FMAX";
   case X86ISD::FMAX_RND:           return "X86ISD::FMAX_RND";
   case X86ISD::FMIN:               return "X86ISD::FMIN";
   case X86ISD::FMIN_RND:           return "X86ISD::FMIN_RND";
   case X86ISD::FMAXC:              return "X86ISD::FMAXC";
   case X86ISD::FMINC:              return "X86ISD::FMINC";
   case X86ISD::FRSQRT:             return "X86ISD::FRSQRT";
+  case X86ISD::FRSQRTS:             return "X86ISD::FRSQRTS";
   case X86ISD::FRCP:               return "X86ISD::FRCP";
+  case X86ISD::FRCPS:              return "X86ISD::FRCPS";
   case X86ISD::EXTRQI:             return "X86ISD::EXTRQI";
   case X86ISD::INSERTQI:           return "X86ISD::INSERTQI";
   case X86ISD::TLSADDR:            return "X86ISD::TLSADDR";
   case X86ISD::TLSBASEADDR:        return "X86ISD::TLSBASEADDR";
   case X86ISD::TLSCALL:            return "X86ISD::TLSCALL";
   case X86ISD::EH_SJLJ_SETJMP:     return "X86ISD::EH_SJLJ_SETJMP";
   case X86ISD::EH_SJLJ_LONGJMP:    return "X86ISD::EH_SJLJ_LONGJMP";
   case X86ISD::EH_RETURN:          return "X86ISD::EH_RETURN";

lib/Target/X86/X86InstrFragmentsSIMD.td

 def X86for     : SDNode<"X86ISD::FOR",       SDTFPBinOp,
                         [SDNPCommutative, SDNPAssociative]>;
 def X86fxor    : SDNode<"X86ISD::FXOR",      SDTFPBinOp,
                         [SDNPCommutative, SDNPAssociative]>;
 def X86fandn   : SDNode<"X86ISD::FANDN",     SDTFPBinOp,
                         [SDNPCommutative, SDNPAssociative]>;
 def X86frsqrt  : SDNode<"X86ISD::FRSQRT",    SDTFPUnaryOp>;
 def X86frcp    : SDNode<"X86ISD::FRCP",      SDTFPUnaryOp>;
-def X86frsqrt14s: SDNode<"X86ISD::FRSQRT",  SDTFPBinOp>;
-def X86frcp14s : SDNode<"X86ISD::FRCP",    SDTFPBinOp>;
+def X86frsqrt14s: SDNode<"X86ISD::FRSQRTS",  SDTFPBinOp>;
+def X86frcp14s : SDNode<"X86ISD::FRCPS",    SDTFPBinOp>;
 def X86fhadd   : SDNode<"X86ISD::FHADD",     SDTFPBinOp>;
 def X86fhsub   : SDNode<"X86ISD::FHSUB",     SDTFPBinOp>;
 def X86hadd    : SDNode<"X86ISD::HADD",      SDTIntBinOp>;
 def X86hsub    : SDNode<"X86ISD::HSUB",      SDTIntBinOp>;
 def X86comi    : SDNode<"X86ISD::COMI",      SDTX86CmpTest>;
 def X86comiSae : SDNode<"X86ISD::COMI",      SDTX86CmpTestSae>;
 def X86ucomi   : SDNode<"X86ISD::UCOMI",     SDTX86CmpTest>;
 def X86ucomiSae: SDNode<"X86ISD::UCOMI",     SDTX86CmpTestSae>;

lib/Target/X86/X86IntrinsicsInfo.h

   X86_INTRINSIC_DATA(avx512_ptestnm_w_256, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_w_512, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_pd_128, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_pd_256, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_pd_512, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_ps_128, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_ps_256, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_ps_512, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
-  X86_INTRINSIC_DATA(avx512_rcp14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::FRCP, 0),
-  X86_INTRINSIC_DATA(avx512_rcp14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::FRCP, 0),
+  X86_INTRINSIC_DATA(avx512_rcp14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::FRCPS, 0),
+  X86_INTRINSIC_DATA(avx512_rcp14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::FRCPS, 0),
   X86_INTRINSIC_DATA(avx512_rcp28_pd, INTR_TYPE_1OP_MASK_RM, X86ISD::RCP28, 0),
   X86_INTRINSIC_DATA(avx512_rcp28_ps, INTR_TYPE_1OP_MASK_RM, X86ISD::RCP28, 0),
   X86_INTRINSIC_DATA(avx512_rcp28_sd, INTR_TYPE_SCALAR_MASK_RM, X86ISD::RCP28, 0),
   X86_INTRINSIC_DATA(avx512_rcp28_ss, INTR_TYPE_SCALAR_MASK_RM, X86ISD::RCP28, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_pd_128, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_pd_256, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_pd_512, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_ps_128, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_ps_256, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt14_ps_512, INTR_TYPE_1OP_MASK, X86ISD::FRSQRT, 0),
-  X86_INTRINSIC_DATA(avx512_rsqrt14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::FRSQRT, 0),
-  X86_INTRINSIC_DATA(avx512_rsqrt14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::FRSQRT, 0),
+  X86_INTRINSIC_DATA(avx512_rsqrt14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::FRSQRTS, 0),
+  X86_INTRINSIC_DATA(avx512_rsqrt14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::FRSQRTS, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt28_pd, INTR_TYPE_1OP_MASK_RM,X86ISD::RSQRT28, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt28_ps, INTR_TYPE_1OP_MASK_RM,X86ISD::RSQRT28, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt28_sd, INTR_TYPE_SCALAR_MASK_RM,X86ISD::RSQRT28, 0),
   X86_INTRINSIC_DATA(avx512_rsqrt28_ss, INTR_TYPE_SCALAR_MASK_RM,X86ISD::RSQRT28, 0),
   X86_INTRINSIC_DATA(avx512_vcomi_sd, COMI_RM, X86ISD::COMI, X86ISD::UCOMI),
   X86_INTRINSIC_DATA(avx512_vcomi_ss, COMI_RM, X86ISD::COMI, X86ISD::UCOMI),
   X86_INTRINSIC_DATA(avx512_vcvtsd2si32, INTR_TYPE_2OP,
                      X86ISD::SCALAR_FP_TO_SINT_RND, 0),

test/CodeGen/X86/avx512-intrinsics.ll

 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vrsqrt14ps %zmm0, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1) ; <<16 x float>> [#uses=1]
   ret <16 x float> %res
 }
 declare <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float>, <16 x float>, i16) nounwind readnone
 
-define <4 x float> @test_rsqrt14_ss(<4 x float> %a0) {
-; CHECK-LABEL: test_rsqrt14_ss:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vrsqrt14ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    retq
-  %res = call <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
-  ret <4 x float> %res
-}
-declare <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
-define <4 x float> @test_rcp14_ss(<4 x float> %a0) {
-; CHECK-LABEL: test_rcp14_ss:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vrcp14ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    retq
-  %res = call <4 x float> @llvm.x86.avx512.rcp14.ss(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
-  ret <4 x float> %res
-}
-declare <4 x float> @llvm.x86.avx512.rcp14.ss(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
 define <8 x double> @test_sqrt_pd_512(<8 x double> %a0) {
 ; CHECK-LABEL: test_sqrt_pd_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vsqrtpd %zmm0, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <8 x double> @llvm.x86.avx512.mask.sqrt.pd.512(<8 x double> %a0,  <8 x double> zeroinitializer, i8 -1, i32 4)
   ret <8 x double> %res
 }

test/CodeGen/X86/avx512-scalarIntrinsics.ll

+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
+
+
+define <4 x float> @test_rsqrt14_ss(<4 x float> %a0) {
+  ; CHECK-LABEL: test_rsqrt14_ss:
+  ; CHECK:       ## BB#0:
+  ; CHECK-NEXT:    vrsqrt14ss %xmm0, %xmm0, %xmm0
+  ; CHECK-NEXT:    retq
+    %res = call <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
+    ret <4 x float> %res
+}
+declare <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_rcp14_ss(<4 x float> %a0) {
+  ; CHECK-LABEL: test_rcp14_ss:
+  ; CHECK:       ## BB#0:
+  ; CHECK-NEXT:    vrcp14ss %xmm0, %xmm0, %xmm0
+  ; CHECK-NEXT:    retq
+    %res = call <4 x float> @llvm.x86.avx512.rcp14.ss(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
+    ret <4 x float> %res
+}
+
+define <4 x float> @test_rsqrt14_sd(<4 x float> %a0) {
+  ; CHECK-LABEL: test_rsqrt14_sd:
+  ; CHECK:       ## BB#0:
+  ; CHECK-NEXT:    vrsqrt14sd %xmm0, %xmm0, %xmm0
+  ; CHECK-NEXT:    retq
+    %res = call <4 x float> @llvm.x86.avx512.rsqrt14.sd(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
+    ret <4 x float> %res
+}
+declare <4 x float> @llvm.x86.avx512.rsqrt14.sd(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_rcp14_sd(<4 x float> %a0) {
+  ; CHECK-LABEL: test_rcp14_sd:
+  ; CHECK:       ## BB#0:
+  ; CHECK-NEXT:    vrcp14sd %xmm0, %xmm0, %xmm0
+  ; CHECK-NEXT:    retq
+    %res = call <4 x float> @llvm.x86.avx512.rcp14.sd(<4 x float> %a0, <4 x float> %a0, <4 x float> zeroinitializer, i8 -1) ; <<4 x float>> [#uses=1]
+    ret <4 x float> %res
+}
+
+