From: Craig Topper <craig.topper@gmail.com>
Date: Sun, 12 Mar 2017 23:05:00 +0000 (+0000)
Subject: [X86] Lower SSE/AVX cmpps/pd intrinsics directly to X86ISD::CMPP SDNodes.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=b853c36a907e895ca4bd2f9a00c34a0218767dd3;p=llvm

[X86] Lower SSE/AVX cmpps/pd intrinsics directly to X86ISD::CMPP SDNodes.

This allows us to remove a duplicate set of patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297593 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 037f0282a91..82ba58d81ce 100644
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -34,9 +34,9 @@ def load_mvmmx : PatFrag<(ops node:$ptr),
 // SSE specific DAG Nodes.
 //===----------------------------------------------------------------------===//
 
-def SDTX86VFCMP : SDTypeProfile<1, 3, [SDTCisFP<0>, SDTCisSameAs<1, 2>,
-                                       SDTCisFP<1>, SDTCisVT<3, i8>,
-                                       SDTCisVec<1>]>;
+def SDTX86VFCMP : SDTypeProfile<1, 3, [SDTCisFP<0>, SDTCisVec<0>,
+                                       SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>,
+                                       SDTCisVT<3, i8>]>;
 def SDTX86CmpTestSae : SDTypeProfile<1, 3, [SDTCisVT<0, i32>, 
                                      SDTCisSameAs<1, 2>, SDTCisInt<3>]>;
 
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index f1281280d90..e5e2e32d68d 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -2460,18 +2460,19 @@ let Defs = [EFLAGS] in {
 
 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
-                            Operand CC, Intrinsic Int, string asm,
+                            Operand CC,  ValueType VT, string asm,
                             string asm_alt, Domain d, ImmLeaf immLeaf,
                             PatFrag ld_frag, OpndItins itins = SSE_ALU_F32P> {
   let isCommutable = 1 in
   def rri : PIi8<0xC2, MRMSrcReg,
              (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
-             [(set RC:$dst, (Int RC:$src1, RC:$src2, immLeaf:$cc))],
+             [(set RC:$dst, (VT (X86cmpp RC:$src1, RC:$src2, immLeaf:$cc)))],
              itins.rr, d>,
             Sched<[WriteFAdd]>;
   def rmi : PIi8<0xC2, MRMSrcMem,
              (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
-             [(set RC:$dst, (Int RC:$src1, (ld_frag addr:$src2), immLeaf:$cc))],
+             [(set RC:$dst,
+               (VT (X86cmpp RC:$src1, (ld_frag addr:$src2), immLeaf:$cc)))],
              itins.rm, d>,
             Sched<[WriteFAddLd, ReadAfterLd]>;
 
@@ -2488,67 +2489,33 @@ multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
   }
 }
 
-defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
+defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, v4f32,
                "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
                SSEPackedSingle, i8immZExt5, loadv4f32>, PS, VEX_4V, VEX_WIG;
-defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
+defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, v2f64,
                "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
                SSEPackedDouble, i8immZExt5, loadv2f64>, PD, VEX_4V, VEX_WIG;
-defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
+defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, v8f32,
                "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
                SSEPackedSingle, i8immZExt5, loadv8f32>, PS, VEX_4V, VEX_L;
-defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
+defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, v4f64,
                "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
                SSEPackedDouble, i8immZExt5, loadv4f64>, PD, VEX_4V, VEX_L;
 let Constraints = "$src1 = $dst" in {
-  defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
+  defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, v4f32,
                  "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
                  "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
                  SSEPackedSingle, i8immZExt5, memopv4f32, SSE_ALU_F32P>, PS;
-  defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
+  defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, v2f64,
                  "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
                  "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
                  SSEPackedDouble, i8immZExt5, memopv2f64, SSE_ALU_F64P>, PD;
 }
 
-let Predicates = [HasAVX] in {
-def : Pat<(v4f32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
-          (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
-def : Pat<(v4f32 (X86cmpp (v4f32 VR128:$src1), (loadv4f32 addr:$src2), imm:$cc)),
-          (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
-def : Pat<(v2f64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
-          (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
-def : Pat<(v2f64 (X86cmpp (v2f64 VR128:$src1), (loadv2f64 addr:$src2), imm:$cc)),
-          (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
-
-def : Pat<(v8f32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
-          (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
-def : Pat<(v8f32 (X86cmpp (v8f32 VR256:$src1), (loadv8f32 addr:$src2), imm:$cc)),
-          (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
-def : Pat<(v4f64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
-          (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
-def : Pat<(v4f64 (X86cmpp (v4f64 VR256:$src1), (loadv4f64 addr:$src2), imm:$cc)),
-          (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
-}
-
-let Predicates = [UseSSE1] in {
-def : Pat<(v4f32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
-          (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
-def : Pat<(v4f32 (X86cmpp (v4f32 VR128:$src1), (memopv4f32 addr:$src2), imm:$cc)),
-          (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
-}
-
-let Predicates = [UseSSE2] in {
-def : Pat<(v2f64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
-          (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
-def : Pat<(v2f64 (X86cmpp (v2f64 VR128:$src1), (memopv2f64 addr:$src2), imm:$cc)),
-          (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
-}
-
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Shuffle Instructions
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/X86/X86IntrinsicsInfo.h b/lib/Target/X86/X86IntrinsicsInfo.h
index c2e950fdbcf..42495bf1e33 100644
--- a/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/lib/Target/X86/X86IntrinsicsInfo.h
@@ -325,6 +325,8 @@ static const IntrinsicData* getIntrinsicWithChain(uint16_t IntNo) {
  * the alphabetical order.
  */
 static const IntrinsicData  IntrinsicsWithoutChain[] = {
+  X86_INTRINSIC_DATA(avx_cmp_pd_256,    INTR_TYPE_3OP, X86ISD::CMPP, 0),
+  X86_INTRINSIC_DATA(avx_cmp_ps_256,    INTR_TYPE_3OP, X86ISD::CMPP, 0),
   X86_INTRINSIC_DATA(avx_cvt_pd2_ps_256,CVTPD2PS, ISD::FP_ROUND, 0),
   X86_INTRINSIC_DATA(avx_cvt_pd2dq_256, INTR_TYPE_1OP, X86ISD::CVTP2SI, 0),
   X86_INTRINSIC_DATA(avx_cvtdq2_ps_256, INTR_TYPE_1OP, ISD::SINT_TO_FP, 0),
@@ -1585,6 +1587,7 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(fma_vfnmsub_pd_256,   INTR_TYPE_3OP, X86ISD::FNMSUB, 0),
   X86_INTRINSIC_DATA(fma_vfnmsub_ps,       INTR_TYPE_3OP, X86ISD::FNMSUB, 0),
   X86_INTRINSIC_DATA(fma_vfnmsub_ps_256,   INTR_TYPE_3OP, X86ISD::FNMSUB, 0),
+  X86_INTRINSIC_DATA(sse_cmp_ps,        INTR_TYPE_3OP, X86ISD::CMPP, 0),
   X86_INTRINSIC_DATA(sse_comieq_ss,     COMI, X86ISD::COMI, ISD::SETEQ),
   X86_INTRINSIC_DATA(sse_comige_ss,     COMI, X86ISD::COMI, ISD::SETGE),
   X86_INTRINSIC_DATA(sse_comigt_ss,     COMI, X86ISD::COMI, ISD::SETGT),
@@ -1605,6 +1608,7 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(sse_ucomile_ss,    COMI, X86ISD::UCOMI, ISD::SETLE),
   X86_INTRINSIC_DATA(sse_ucomilt_ss,    COMI, X86ISD::UCOMI, ISD::SETLT),
   X86_INTRINSIC_DATA(sse_ucomineq_ss,   COMI, X86ISD::UCOMI, ISD::SETNE),
+  X86_INTRINSIC_DATA(sse2_cmp_pd,       INTR_TYPE_3OP, X86ISD::CMPP, 0),
   X86_INTRINSIC_DATA(sse2_comieq_sd,    COMI, X86ISD::COMI, ISD::SETEQ),
   X86_INTRINSIC_DATA(sse2_comige_sd,    COMI, X86ISD::COMI, ISD::SETGE),
   X86_INTRINSIC_DATA(sse2_comigt_sd,    COMI, X86ISD::COMI, ISD::SETGT),