//====- X86InstrMMX.td - Describe the X86 Instruction Set --*- tablegen -*-===//
-//
+//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
-//
+//
//===----------------------------------------------------------------------===//
//
// This file describes the X86 MMX instruction set, defining the instructions,
// MMX Multiclasses
//===----------------------------------------------------------------------===//
-let isTwoAddress = 1 in {
+let Constraints = "$src1 = $dst" in {
// MMXI_binop_rm - Simple MMX binary operator.
multiclass MMXI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType OpVT, bit Commutable = 0> {
- def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
+ def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
+ (ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (OpVT (OpNode VR64:$src1, VR64:$src2)))]> {
let isCommutable = Commutable;
}
- def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), (ins VR64:$src1, i64mem:$src2),
+ def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
+ (ins VR64:$src1, i64mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (OpVT (OpNode VR64:$src1,
(bitconvert
multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
bit Commutable = 0> {
- def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
+ def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
+ (ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))]> {
let isCommutable = Commutable;
}
- def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), (ins VR64:$src1, i64mem:$src2),
+ def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
+ (ins VR64:$src1, i64mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1,
(bitconvert (load_mmx addr:$src2))))]>;
// MMX EMMS & FEMMS Instructions
//===----------------------------------------------------------------------===//
-def MMX_EMMS : MMXI<0x77, RawFrm, (outs), (ins), "emms", [(int_x86_mmx_emms)]>;
-def MMX_FEMMS : MMXI<0x0E, RawFrm, (outs), (ins), "femms", [(int_x86_mmx_femms)]>;
+def MMX_EMMS : MMXI<0x77, RawFrm, (outs), (ins), "emms",
+ [(int_x86_mmx_emms)]>;
+def MMX_FEMMS : MMXI<0x0E, RawFrm, (outs), (ins), "femms",
+ [(int_x86_mmx_femms)]>;
//===----------------------------------------------------------------------===//
// MMX Scalar Instructions
// Data Transfer Instructions
def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (v2i32 (scalar_to_vector GR32:$src)))]>;
+ [(set VR64:$dst,
+ (v2i32 (scalar_to_vector GR32:$src)))]>;
let canFoldAsLoad = 1, isReMaterializable = 1 in
def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (v2i32 (scalar_to_vector (loadi32 addr:$src))))]>;
-let mayStore = 1 in
+ [(set VR64:$dst,
+ (v2i32 (scalar_to_vector (loadi32 addr:$src))))]>;
+let mayStore = 1 in
def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
"movd\t{$src, $dst|$dst, $src}", []>;
[]>;
let neverHasSideEffects = 1 in
-def MMX_MOVD64from64rr : MMXRI<0x7E, MRMSrcReg,
+// These are 64 bit moves, but since the OS X assembler doesn't
+// recognize a register-register movq, we write them as
+// movd.
+def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
(outs GR64:$dst), (ins VR64:$src),
"movd\t{$src, $dst|$dst, $src}", []>;
+def MMX_MOVD64rrv164 : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set VR64:$dst,
+ (v1i64 (scalar_to_vector GR64:$src)))]>;
let neverHasSideEffects = 1 in
def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (v1i64 VR64:$src), addr:$dst)]>;
-def MMX_MOVDQ2Qrr : SDIi8<0xD6, MRMDestMem, (outs VR64:$dst), (ins VR128:$src),
+def MMX_MOVDQ2Qrr : SDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"movdq2q\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(v1i64 (bitconvert
(i64 (vector_extract (v2i64 VR128:$src),
(iPTR 0))))))]>;
-def MMX_MOVQ2DQrr : SSDIi8<0xD6, MRMDestMem, (outs VR128:$dst), (ins VR64:$src),
+def MMX_MOVQ2DQrr : SSDIi8<0xD6, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
"movq2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(movl immAllZerosV,
(v2i64 (scalar_to_vector (i64 (bitconvert VR64:$src))))))]>;
let neverHasSideEffects = 1 in
-def MMX_MOVQ2FR64rr: SSDIi8<0xD6, MRMDestMem, (outs FR64:$dst), (ins VR64:$src),
+def MMX_MOVQ2FR64rr: SSDIi8<0xD6, MRMSrcReg, (outs FR64:$dst), (ins VR64:$src),
"movq2dq\t{$src, $dst|$dst, $src}", []>;
def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
[(set VR64:$dst,
(v2i32 (X86vzmovl (v2i32 (scalar_to_vector GR32:$src)))))]>;
let AddedComplexity = 20 in
-def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
+def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
+ (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(v2i32 (X86vzmovl (v2i32
defm MMX_POR : MMXI_binop_rm_v1i64<0xEB, "por" , or, 1>;
defm MMX_PXOR : MMXI_binop_rm_v1i64<0xEF, "pxor", xor, 1>;
-let isTwoAddress = 1 in {
+let Constraints = "$src1 = $dst" in {
def MMX_PANDNrr : MMXI<0xDF, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
"pandn\t{$src2, $dst|$dst, $src2}",
// Conversion Instructions
// -- Unpack Instructions
-let isTwoAddress = 1 in {
+let Constraints = "$src1 = $dst" in {
// Unpack High Packed Data Instructions
- def MMX_PUNPCKHBWrr : MMXI<0x68, MRMSrcReg,
+ def MMX_PUNPCKHBWrr : MMXI<0x68, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
"punpckhbw\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v8i8 (mmx_unpckh VR64:$src1, VR64:$src2)))]>;
- def MMX_PUNPCKHBWrm : MMXI<0x68, MRMSrcMem,
+ def MMX_PUNPCKHBWrm : MMXI<0x68, MRMSrcMem,
(outs VR64:$dst), (ins VR64:$src1, i64mem:$src2),
"punpckhbw\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v8i8 (mmx_unpckh VR64:$src1,
(bc_v8i8 (load_mmx addr:$src2)))))]>;
- def MMX_PUNPCKHWDrr : MMXI<0x69, MRMSrcReg,
+ def MMX_PUNPCKHWDrr : MMXI<0x69, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
"punpckhwd\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v4i16 (mmx_unpckh VR64:$src1, VR64:$src2)))]>;
- def MMX_PUNPCKHWDrm : MMXI<0x69, MRMSrcMem,
+ def MMX_PUNPCKHWDrm : MMXI<0x69, MRMSrcMem,
(outs VR64:$dst), (ins VR64:$src1, i64mem:$src2),
"punpckhwd\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v4i16 (mmx_unpckh VR64:$src1,
(bc_v4i16 (load_mmx addr:$src2)))))]>;
- def MMX_PUNPCKHDQrr : MMXI<0x6A, MRMSrcReg,
+ def MMX_PUNPCKHDQrr : MMXI<0x6A, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
"punpckhdq\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v4i16 (mmx_unpckl VR64:$src1,
(bc_v4i16 (load_mmx addr:$src2)))))]>;
- def MMX_PUNPCKLDQrr : MMXI<0x62, MRMSrcReg,
+ def MMX_PUNPCKLDQrr : MMXI<0x62, MRMSrcReg,
(outs VR64:$dst), (ins VR64:$src1, VR64:$src2),
"punpckldq\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
(v2i32 (mmx_unpckl VR64:$src1, VR64:$src2)))]>;
- def MMX_PUNPCKLDQrm : MMXI<0x62, MRMSrcMem,
+ def MMX_PUNPCKLDQrm : MMXI<0x62, MRMSrcMem,
(outs VR64:$dst), (ins VR64:$src1, i64mem:$src2),
"punpckldq\t{$src2, $dst|$dst, $src2}",
[(set VR64:$dst,
def MMX_CVTPD2PIrr : MMX2I<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"cvtpd2pi\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
-def MMX_CVTPD2PIrm : MMX2I<0x2D, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
+def MMX_CVTPD2PIrm : MMX2I<0x2D, MRMSrcMem, (outs VR64:$dst),
+ (ins f128mem:$src),
"cvtpd2pi\t{$src, $dst|$dst, $src}", []>;
def MMX_CVTPI2PDrr : MMX2I<0x2A, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
"cvtpi2pd\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
-def MMX_CVTPI2PDrm : MMX2I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
+def MMX_CVTPI2PDrm : MMX2I<0x2A, MRMSrcMem, (outs VR128:$dst),
+ (ins i64mem:$src),
"cvtpi2pd\t{$src, $dst|$dst, $src}", []>;
def MMX_CVTPI2PSrr : MMXI<0x2A, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
"cvtpi2ps\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
-def MMX_CVTPI2PSrm : MMXI<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
+def MMX_CVTPI2PSrm : MMXI<0x2A, MRMSrcMem, (outs VR128:$dst),
+ (ins i64mem:$src),
"cvtpi2ps\t{$src, $dst|$dst, $src}", []>;
def MMX_CVTPS2PIrr : MMXI<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
def MMX_CVTTPD2PIrr : MMX2I<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"cvttpd2pi\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
-def MMX_CVTTPD2PIrm : MMX2I<0x2C, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
+def MMX_CVTTPD2PIrm : MMX2I<0x2C, MRMSrcMem, (outs VR64:$dst),
+ (ins f128mem:$src),
"cvttpd2pi\t{$src, $dst|$dst, $src}", []>;
def MMX_CVTTPS2PIrr : MMXI<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set GR32:$dst, (MMX_X86pextrw (v4i16 VR64:$src1),
(iPTR imm:$src2)))]>;
-let isTwoAddress = 1 in {
+let Constraints = "$src1 = $dst" in {
def MMX_PINSRWrri : MMXIi8<0xC4, MRMSrcReg,
- (outs VR64:$dst), (ins VR64:$src1, GR32:$src2, i16i8imm:$src3),
+ (outs VR64:$dst), (ins VR64:$src1, GR32:$src2,
+ i16i8imm:$src3),
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR64:$dst, (v4i16 (MMX_X86pinsrw (v4i16 VR64:$src1),
- GR32:$src2, (iPTR imm:$src3))))]>;
+ GR32:$src2,(iPTR imm:$src3))))]>;
def MMX_PINSRWrmi : MMXIi8<0xC4, MRMSrcMem,
- (outs VR64:$dst), (ins VR64:$src1, i16mem:$src2, i16i8imm:$src3),
+ (outs VR64:$dst), (ins VR64:$src1, i16mem:$src2,
+ i16i8imm:$src3),
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR64:$dst,
(v4i16 (MMX_X86pinsrw (v4i16 VR64:$src1),
// Misc.
let Uses = [EDI] in
-def MMX_MASKMOVQ : MMXI<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
+def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)]>;
let Uses = [RDI] in
-def MMX_MASKMOVQ64: MMXI64<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
+def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)]>;
//===----------------------------------------------------------------------===//
// Alias instructions that map zero vector to pxor.
-let isReMaterializable = 1 in {
+let isReMaterializable = 1, isCodeGenOnly = 1 in {
def MMX_V_SET0 : MMXI<0xEF, MRMInitReg, (outs VR64:$dst), (ins),
"pxor\t$dst, $dst",
[(set VR64:$dst, (v2i32 immAllZerosV))]>;
def : Pat<(f64 (bitconvert (v8i8 VR64:$src))),
(MMX_MOVQ2FR64rr VR64:$src)>;
-// Move scalar to MMX zero-extended
-// movd to MMX register zero-extends
-let AddedComplexity = 15 in {
- def : Pat<(v8i8 (X86vzmovl (bc_v8i8 (v2i32 (scalar_to_vector GR32:$src))))),
- (MMX_MOVZDI2PDIrr GR32:$src)>;
- def : Pat<(v4i16 (X86vzmovl (bc_v4i16 (v2i32 (scalar_to_vector GR32:$src))))),
- (MMX_MOVZDI2PDIrr GR32:$src)>;
-}
-
let AddedComplexity = 20 in {
- def : Pat<(v8i8 (X86vzmovl (bc_v8i8 (load_mmx addr:$src)))),
- (MMX_MOVZDI2PDIrm addr:$src)>;
- def : Pat<(v4i16 (X86vzmovl (bc_v4i16 (load_mmx addr:$src)))),
- (MMX_MOVZDI2PDIrm addr:$src)>;
def : Pat<(v2i32 (X86vzmovl (bc_v2i32 (load_mmx addr:$src)))),
- (MMX_MOVZDI2PDIrm addr:$src)>;
+ (MMX_MOVZDI2PDIrm addr:$src)>;
}
// Clear top half.
let AddedComplexity = 15 in {
- def : Pat<(v8i8 (X86vzmovl VR64:$src)),
- (MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
- def : Pat<(v4i16 (X86vzmovl VR64:$src)),
- (MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
def : Pat<(v2i32 (X86vzmovl VR64:$src)),
(MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
}
-// Scalar to v4i16 / v8i8. The source may be a GR32, but only the lower
-// 8 or 16-bits matter.
-def : Pat<(bc_v8i8 (v2i32 (scalar_to_vector GR32:$src))),
- (MMX_MOVD64rr GR32:$src)>;
-def : Pat<(bc_v4i16 (v2i32 (scalar_to_vector GR32:$src))),
- (MMX_MOVD64rr GR32:$src)>;
-
// Patterns to perform canonical versions of vector shuffling.
let AddedComplexity = 10 in {
def : Pat<(v8i8 (mmx_unpckl_undef VR64:$src, (undef))),
(iPTR 0))))),
(v8i8 (MMX_MOVDQ2Qrr VR128:$src))>;
+// Patterns for vector comparisons
+def : Pat<(v8i8 (X86pcmpeqb VR64:$src1, VR64:$src2)),
+ (MMX_PCMPEQBrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v8i8 (X86pcmpeqb VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPEQBrm VR64:$src1, addr:$src2)>;
+def : Pat<(v4i16 (X86pcmpeqw VR64:$src1, VR64:$src2)),
+ (MMX_PCMPEQWrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v4i16 (X86pcmpeqw VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPEQWrm VR64:$src1, addr:$src2)>;
+def : Pat<(v2i32 (X86pcmpeqd VR64:$src1, VR64:$src2)),
+ (MMX_PCMPEQDrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v2i32 (X86pcmpeqd VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPEQDrm VR64:$src1, addr:$src2)>;
+
+def : Pat<(v8i8 (X86pcmpgtb VR64:$src1, VR64:$src2)),
+ (MMX_PCMPGTBrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v8i8 (X86pcmpgtb VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPGTBrm VR64:$src1, addr:$src2)>;
+def : Pat<(v4i16 (X86pcmpgtw VR64:$src1, VR64:$src2)),
+ (MMX_PCMPGTWrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v4i16 (X86pcmpgtw VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPGTWrm VR64:$src1, addr:$src2)>;
+def : Pat<(v2i32 (X86pcmpgtd VR64:$src1, VR64:$src2)),
+ (MMX_PCMPGTDrr VR64:$src1, VR64:$src2)>;
+def : Pat<(v2i32 (X86pcmpgtd VR64:$src1, (bitconvert (load_mmx addr:$src2)))),
+ (MMX_PCMPGTDrm VR64:$src1, addr:$src2)>;
+
// CMOV* - Used to implement the SELECT DAG operation. Expanded by the
// scheduler into a branch sequence.
// These are expanded by the scheduler.