aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/X86/X86InstrMMX.td
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/X86/X86InstrMMX.td')
-rw-r--r--lib/Target/X86/X86InstrMMX.td80
1 files changed, 54 insertions, 26 deletions
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index 127af6f7f9..49721df7c1 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -20,6 +20,7 @@
// MMX Multiclasses
//===----------------------------------------------------------------------===//
+let Sched = WriteVecALU in {
def MMX_INTALU_ITINS : OpndItins<
IIC_MMX_ALU_RR, IIC_MMX_ALU_RM
>;
@@ -35,11 +36,14 @@ def MMX_PHADDSUBW : OpndItins<
def MMX_PHADDSUBD : OpndItins<
IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM
>;
+}
+let Sched = WriteVecIMul in
def MMX_PMUL_ITINS : OpndItins<
IIC_MMX_PMUL, IIC_MMX_PMUL
>;
+let Sched = WriteVecALU in {
def MMX_PSADBW_ITINS : OpndItins<
IIC_MMX_PSADBW, IIC_MMX_PSADBW
>;
@@ -47,11 +51,13 @@ def MMX_PSADBW_ITINS : OpndItins<
def MMX_MISC_FUNC_ITINS : OpndItins<
IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG
>;
+}
def MMX_SHIFT_ITINS : ShiftOpndItins<
IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI
>;
+let Sched = WriteShuffle in {
def MMX_UNPCK_H_ITINS : OpndItins<
IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM
>;
@@ -67,7 +73,9 @@ def MMX_PCK_ITINS : OpndItins<
def MMX_PSHUF_ITINS : OpndItins<
IIC_MMX_PSHUF, IIC_MMX_PSHUF
>;
+} // Sched
+let Sched = WriteCvtF2I in {
def MMX_CVT_PD_ITINS : OpndItins<
IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM
>;
@@ -75,6 +83,7 @@ def MMX_CVT_PD_ITINS : OpndItins<
def MMX_CVT_PS_ITINS : OpndItins<
IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM
>;
+}
let Constraints = "$src1 = $dst" in {
// MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic.
@@ -84,7 +93,8 @@ let Constraints = "$src1 = $dst" in {
def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr> {
+ [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
+ Sched<[itins.Sched]> {
let isCommutable = Commutable;
}
def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
@@ -92,7 +102,7 @@ let Constraints = "$src1 = $dst" in {
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1,
(bitconvert (load_mmx addr:$src2))))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
@@ -101,17 +111,19 @@ let Constraints = "$src1 = $dst" in {
def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>;
+ [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
+ Sched<[WriteVecShift]>;
def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
(ins VR64:$src1, i64mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst, (IntId VR64:$src1,
(bitconvert (load_mmx addr:$src2))))],
- itins.rm>;
+ itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>;
def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst),
(ins VR64:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>;
+ [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>,
+ Sched<[WriteVecShift]>;
}
}
@@ -120,13 +132,14 @@ multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
Intrinsic IntId64, OpndItins itins> {
def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>;
+ [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR64:$dst,
(IntId64 (bitconvert (memopmmx addr:$src))))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded]>;
}
/// Binary MMX instructions requiring SSSE3.
@@ -137,13 +150,15 @@ multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>;
+ [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst),
(ins VR64:$src1, i64mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR64:$dst,
(IntId64 VR64:$src1,
- (bitconvert (memopmmx addr:$src2))))], itins.rm>;
+ (bitconvert (memopmmx addr:$src2))))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
@@ -164,9 +179,11 @@ multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
string asm, OpndItins itins, Domain d> {
def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
- [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>;
+ [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>,
+ Sched<[itins.Sched]>;
def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
- [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>;
+ [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>,
+ Sched<[itins.Sched.Folded]>;
}
multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
@@ -174,11 +191,11 @@ multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
PatFrag ld_frag, string asm, Domain d> {
def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst),(ins DstRC:$src1, SrcRC:$src2),
asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
- IIC_DEFAULT, d>;
+ NoItinerary, d>;
def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
(ins DstRC:$src1, x86memop:$src2), asm,
[(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
- IIC_DEFAULT, d>;
+ NoItinerary, d>;
}
//===----------------------------------------------------------------------===//
@@ -197,16 +214,17 @@ def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (scalar_to_vector GR32:$src)))],
- IIC_MMX_MOV_MM_RM>;
+ IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
let canFoldAsLoad = 1 in
def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (scalar_to_vector (loadi32 addr:$src))))],
- IIC_MMX_MOV_MM_RM>;
+ IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
let mayStore = 1 in
def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
- "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>;
+ "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>,
+ Sched<[WriteStore]>;
// Low word of MMX to GPR.
def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1,
@@ -214,16 +232,18 @@ def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1,
def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst,
- (MMX_X86movd2w (x86mmx VR64:$src)))], IIC_MMX_MOV_REG_MM>;
+ (MMX_X86movd2w (x86mmx VR64:$src)))],
+ IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>;
let neverHasSideEffects = 1 in
def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
"movd\t{$src, $dst|$dst, $src}",
- [], IIC_MMX_MOV_MM_RM>;
+ [], IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
// These are 64 bit moves, but since the OS X assembler doesn't
// recognize a register-register movq, we write them as
// movd.
+let SchedRW = [WriteMove] in {
def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
(outs GR64:$dst), (ins VR64:$src),
"movd\t{$src, $dst|$dst, $src}",
@@ -237,6 +257,9 @@ let neverHasSideEffects = 1 in
def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}", [],
IIC_MMX_MOVQ_RR>;
+} // SchedRW
+
+let SchedRW = [WriteLoad] in {
let canFoldAsLoad = 1 in
def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
"movq\t{$src, $dst|$dst, $src}",
@@ -246,7 +269,9 @@ def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (x86mmx VR64:$src), addr:$dst)],
IIC_MMX_MOVQ_RM>;
+} // SchedRW
+let SchedRW = [WriteMove] in {
def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
(ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
@@ -271,11 +296,12 @@ def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
(ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}",
[], IIC_MMX_MOVQ_RR>;
+} // SchedRW
def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
"movntq\t{$src, $dst|$dst, $src}",
[(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)],
- IIC_MMX_MOVQ_RM>;
+ IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>;
let AddedComplexity = 15 in
// movd to MMX register zero-extends
@@ -283,7 +309,7 @@ def MMX_MOVZDI2PDIrr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))))],
- IIC_MMX_MOV_MM_RM>;
+ IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
let AddedComplexity = 20 in
def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
(ins i32mem:$src),
@@ -291,7 +317,7 @@ def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
[(set VR64:$dst,
(x86mmx (X86vzmovl (x86mmx
(scalar_to_vector (loadi32 addr:$src))))))],
- IIC_MMX_MOV_MM_RM>;
+ IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
// Arithmetic Instructions
defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b,
@@ -491,14 +517,14 @@ def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg,
"pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR64:$dst,
(int_x86_sse_pshuf_w VR64:$src1, imm:$src2))],
- IIC_MMX_PSHUF>;
+ IIC_MMX_PSHUF>, Sched<[WriteShuffle]>;
def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
(outs VR64:$dst), (ins i64mem:$src1, i8imm:$src2),
"pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR64:$dst,
(int_x86_sse_pshuf_w (load_mmx addr:$src1),
imm:$src2))],
- IIC_MMX_PSHUF>;
+ IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>;
@@ -532,7 +558,7 @@ def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg,
"pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set GR32:$dst, (int_x86_mmx_pextr_w VR64:$src1,
(iPTR imm:$src2)))],
- IIC_MMX_PEXTR>;
+ IIC_MMX_PEXTR>, Sched<[WriteShuffle]>;
let Constraints = "$src1 = $dst" in {
def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg,
(outs VR64:$dst),
@@ -540,7 +566,7 @@ let Constraints = "$src1 = $dst" in {
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
GR32:$src2, (iPTR imm:$src3)))],
- IIC_MMX_PINSRW>;
+ IIC_MMX_PINSRW>, Sched<[WriteShuffle]>;
def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem,
(outs VR64:$dst),
@@ -549,7 +575,7 @@ let Constraints = "$src1 = $dst" in {
[(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
(i32 (anyext (loadi16 addr:$src2))),
(iPTR imm:$src3)))],
- IIC_MMX_PINSRW>;
+ IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
// Mask creation
@@ -570,6 +596,7 @@ def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))),
(x86mmx (MMX_MOVQ64rm addr:$src))>;
// Misc.
+let SchedRW = [WriteShuffle] in {
let Uses = [EDI] in
def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
@@ -580,6 +607,7 @@ def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)],
IIC_MMX_MASKMOV>;
+}
// 64-bit bit convert.
let Predicates = [HasSSE2] in {
generated by cgit v1.2.3-18-g5258 (git 2.32.0) at 2024-11-21 12:32:03 +0000