From fdb8caab02f13bfd53979b51e3a3780d8d018e8a Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Sat, 31 Aug 2019 23:52:25 +0000 Subject: [PATCH] [X86] Replace some COPY_TO_REGCLASS from GR32/GR64 to VR128 in isel patterns with VMOVDI2PDIrr/VMOV64toPQIrr. This is what the copies will eventually be turned into. We don't use COPY_TO_REGCLASS for scalar_to_vector patterns. So we should use the real instruction here too. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370601 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/X86/X86InstrSSE.td | 40 ++++++++++++++++------------------- 1 file changed, 18 insertions(+), 22 deletions(-) diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 1e2703aa427..16c76ca7a6e 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -7469,36 +7469,32 @@ let Predicates = [HasAVX2, NoVLX] in { let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in { def : Pat<(v16i8 (X86VBroadcast GR8:$src)), - (VPBROADCASTBrr (v16i8 (COPY_TO_REGCLASS + (VPBROADCASTBrr (VMOVDI2PDIrr (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)), - GR8:$src, sub_8bit)), - VR128)))>; + GR8:$src, sub_8bit))))>; def : Pat<(v32i8 (X86VBroadcast GR8:$src)), - (VPBROADCASTBYrr (v16i8 (COPY_TO_REGCLASS + (VPBROADCASTBYrr (VMOVDI2PDIrr (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)), - GR8:$src, sub_8bit)), - VR128)))>; + GR8:$src, sub_8bit))))>; def : Pat<(v8i16 (X86VBroadcast GR16:$src)), - (VPBROADCASTWrr (v8i16 (COPY_TO_REGCLASS + (VPBROADCASTWrr (VMOVDI2PDIrr (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)), - GR16:$src, sub_16bit)), - VR128)))>; + GR16:$src, sub_16bit))))>; def : Pat<(v16i16 (X86VBroadcast GR16:$src)), - (VPBROADCASTWYrr (v8i16 (COPY_TO_REGCLASS + (VPBROADCASTWYrr (VMOVDI2PDIrr (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)), - GR16:$src, sub_16bit)), - VR128)))>; + GR16:$src, sub_16bit))))>; } let Predicates = [HasAVX2, NoVLX] in { def : Pat<(v4i32 (X86VBroadcast GR32:$src)), - (VPBROADCASTDrr (v4i32 (COPY_TO_REGCLASS GR32:$src, VR128)))>; + (VPBROADCASTDrr (VMOVDI2PDIrr GR32:$src))>; def : Pat<(v8i32 (X86VBroadcast GR32:$src)), - (VPBROADCASTDYrr (v4i32 (COPY_TO_REGCLASS GR32:$src, VR128)))>; + (VPBROADCASTDYrr (VMOVDI2PDIrr GR32:$src))>; def : Pat<(v2i64 (X86VBroadcast GR64:$src)), - (VPBROADCASTQrr (v2i64 (COPY_TO_REGCLASS GR64:$src, VR128)))>; + (VPBROADCASTQrr (VMOV64toPQIrr GR64:$src))>; def : Pat<(v4i64 (X86VBroadcast GR64:$src)), - (VPBROADCASTQYrr (v2i64 (COPY_TO_REGCLASS GR64:$src, VR128)))>; + (VPBROADCASTQYrr (VMOV64toPQIrr GR64:$src))>; } // AVX1 broadcast patterns @@ -7541,18 +7537,18 @@ let Predicates = [HasAVX1Only] in { (v2f64 (VMOVDDUPrr (v2f64 (COPY_TO_REGCLASS FR64:$src, VR128)))), 1)>; def : Pat<(v4i32 (X86VBroadcast GR32:$src)), - (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR32:$src, VR128)), 0)>; + (VPSHUFDri (VMOVDI2PDIrr GR32:$src), 0)>; def : Pat<(v8i32 (X86VBroadcast GR32:$src)), (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), - (v4i32 (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR32:$src, VR128)), 0)), sub_xmm), - (v4i32 (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR32:$src, VR128)), 0)), 1)>; + (v4i32 (VPSHUFDri (VMOVDI2PDIrr GR32:$src), 0)), sub_xmm), + (v4i32 (VPSHUFDri (VMOVDI2PDIrr GR32:$src), 0)), 1)>; def : Pat<(v4i64 (X86VBroadcast GR64:$src)), (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), - (v4i32 (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR64:$src, VR128)), 0x44)), sub_xmm), - (v4i32 (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR64:$src, VR128)), 0x44)), 1)>; + (v4i32 (VPSHUFDri (VMOV64toPQIrr GR64:$src), 0x44)), sub_xmm), + (v4i32 (VPSHUFDri (VMOV64toPQIrr GR64:$src), 0x44)), 1)>; def : Pat<(v2i64 (X86VBroadcast i64:$src)), - (VPSHUFDri (v4i32 (COPY_TO_REGCLASS GR64:$src, VR128)), 0x44)>; + (VPSHUFDri (VMOV64toPQIrr GR64:$src), 0x44)>; def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))), (VMOVDDUPrm addr:$src)>; } -- 2.50.1