OSchip
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
13
0
Fork 6
Code Issues Packages Projects Releases Wiki Activity

[X86] Give priority to EVEX FMA instructions over FMA4 instructions. 

No existing processor has both so it doesn't really matter what we do here. But we were previously just relying on pattern order which gave FMA4 priority.

llvm-svn: 317775
This commit is contained in:
Craig Topper 2017-11-09 08:26:26 +00:00
parent 66f32fc431
commit 5bfa5ffe5e
3 changed files with 69 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llvm/lib/Target/X86/X86InstrAVX512.td
View File

@ -6002,12 +6002,12 @@ let Constraints = "$src1 = $dst", hasSideEffects = 0 in {
AVX512FMA3Base, EVEX_B, EVEX_RC;
let isCodeGenOnly = 1, isCommutable = 1 in {
def r : AVX512FMA3<opc, MRMSrcReg, (outs _.FRC:$dst),
def r : AVX512FMA3S<opc, MRMSrcReg, (outs _.FRC:$dst),
(ins _.FRC:$src1, _.FRC:$src2, _.FRC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!if(MaskOnlyReg, [], [RHS_r])>;
def m : AVX512FMA3<opc, MRMSrcMem, (outs _.FRC:$dst),
def m : AVX512FMA3S<opc, MRMSrcMem, (outs _.FRC:$dst),
(ins _.FRC:$src1, _.FRC:$src2, _.ScalarMemOp:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),

34
llvm/lib/Target/X86/X86InstrFMA.td
View File

@ -168,14 +168,14 @@ let ExeDomain = SSEPackedDouble in {
multiclass fma3s_rm_213<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, RegisterClass RC,
SDPatternOperator OpNode> {
def r : FMA3<opc, MRMSrcReg, (outs RC:$dst),
def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set RC:$dst, (OpNode RC:$src2, RC:$src1, RC:$src3))]>;
let mayLoad = 1 in
def m : FMA3<opc, MRMSrcMem, (outs RC:$dst),
def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, RC:$src2, x86memop:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
@ -187,14 +187,14 @@ multiclass fma3s_rm_231<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, RegisterClass RC,
SDPatternOperator OpNode> {
let hasSideEffects = 0 in
def r : FMA3<opc, MRMSrcReg, (outs RC:$dst),
def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[]>;
let mayLoad = 1 in
def m : FMA3<opc, MRMSrcMem, (outs RC:$dst),
def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, RC:$src2, x86memop:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
@ -206,7 +206,7 @@ multiclass fma3s_rm_132<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, RegisterClass RC,
SDPatternOperator OpNode> {
let hasSideEffects = 0 in
def r : FMA3<opc, MRMSrcReg, (outs RC:$dst),
def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
@ -215,7 +215,7 @@ multiclass fma3s_rm_132<bits<8> opc, string OpcodeStr,
// Pattern is 312 order so that the load is in a different place from the
// 213 and 231 patterns this helps tablegen's duplicate pattern detection.
let mayLoad = 1 in
def m : FMA3<opc, MRMSrcMem, (outs RC:$dst),
def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, RC:$src2, x86memop:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
@ -228,7 +228,6 @@ multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
string OpStr, string PackTy, string Suff,
SDNode OpNode, RegisterClass RC,
X86MemOperand x86memop> {
let Predicates = [HasFMA, NoAVX512] in {
defm NAME#213#Suff : fma3s_rm_213<opc213, !strconcat(OpStr, "213", PackTy),
x86memop, RC, OpNode>;
defm NAME#231#Suff : fma3s_rm_231<opc231, !strconcat(OpStr, "231", PackTy),
@ -236,7 +235,6 @@ multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
defm NAME#132#Suff : fma3s_rm_132<opc132, !strconcat(OpStr, "132", PackTy),
x86memop, RC, OpNode>;
}
}
// These FMA*_Int instructions are defined specially for being used when
// the scalar FMA intrinsics are lowered to machine instructions, and in that
@ -255,14 +253,14 @@ let Constraints = "$src1 = $dst", isCommutable = 1, isCodeGenOnly = 1,
hasSideEffects = 0 in
multiclass fma3s_rm_int<bits<8> opc, string OpcodeStr,
Operand memopr, RegisterClass RC> {
def r_Int : FMA3<opc, MRMSrcReg, (outs RC:$dst),
def r_Int : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[]>;
let mayLoad = 1 in
def m_Int : FMA3<opc, MRMSrcMem, (outs RC:$dst),
def m_Int : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, RC:$src2, memopr:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
@ -357,19 +355,19 @@ multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
X86MemOperand x86memop, ValueType OpVT, SDNode OpNode,
PatFrag mem_frag> {
let isCommutable = 1 in
def rr : FMA4<opc, MRMSrcRegOp4, (outs RC:$dst),
def rr : FMA4S<opc, MRMSrcRegOp4, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst,
(OpVT (OpNode RC:$src1, RC:$src2, RC:$src3)))]>, VEX_W, VEX_LIG;
def rm : FMA4<opc, MRMSrcMemOp4, (outs RC:$dst),
def rm : FMA4S<opc, MRMSrcMemOp4, (outs RC:$dst),
(ins RC:$src1, RC:$src2, x86memop:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (OpNode RC:$src1, RC:$src2,
(mem_frag addr:$src3)))]>, VEX_W, VEX_LIG;
def mr : FMA4<opc, MRMSrcMem, (outs RC:$dst),
def mr : FMA4S<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
@ -377,7 +375,7 @@ multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
(OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3))]>, VEX_LIG;
// For disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
def rr_REV : FMA4<opc, MRMSrcReg, (outs RC:$dst),
def rr_REV : FMA4S<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
@ -387,20 +385,20 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
ValueType VT, ComplexPattern mem_cpat, SDNode OpNode> {
let isCodeGenOnly = 1 in {
def rr_Int : FMA4<opc, MRMSrcRegOp4, (outs VR128:$dst),
def rr_Int : FMA4S<opc, MRMSrcRegOp4, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, VR128:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst,
(VT (OpNode VR128:$src1, VR128:$src2, VR128:$src3)))]>, VEX_W,
VEX_LIG;
def rm_Int : FMA4<opc, MRMSrcMemOp4, (outs VR128:$dst),
def rm_Int : FMA4S<opc, MRMSrcMemOp4, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, memop:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst, (VT (OpNode VR128:$src1, VR128:$src2,
mem_cpat:$src3)))]>, VEX_W, VEX_LIG;
def mr_Int : FMA4<opc, MRMSrcMem, (outs VR128:$dst),
def mr_Int : FMA4S<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, memop:$src2, VR128:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
@ -408,7 +406,7 @@ let isCodeGenOnly = 1 in {
(VT (OpNode VR128:$src1, mem_cpat:$src2, VR128:$src3)))]>,
VEX_LIG;
let hasSideEffects = 0 in
def rr_Int_REV : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
def rr_Int_REV : FMA4S<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, VR128:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),

12
llvm/lib/Target/X86/X86InstrFormats.td
View File

@ -827,7 +827,7 @@ class AVX512PIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
class AVX512PI<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag> pattern, Domain d, InstrItinClass itin = NoItinerary>
: I<o, F, outs, ins, asm, pattern, itin, d>, Requires<[HasAVX512]>;
class AVX512FMA3<bits<8> o, Format F, dag outs, dag ins, string asm,
class AVX512FMA3S<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag>pattern, InstrItinClass itin = NoItinerary>
: I<o, F, outs, ins, asm, pattern, itin>, T8PD,
EVEX_4V, Requires<[HasAVX512]>;
@ -867,12 +867,20 @@ class FMA3<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag>pattern, InstrItinClass itin = NoItinerary>
: I<o, F, outs, ins, asm, pattern, itin>, T8PD,
VEX_4V, FMASC, Requires<[HasFMA, NoVLX]>;
class FMA3S<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag>pattern, InstrItinClass itin = NoItinerary>
: I<o, F, outs, ins, asm, pattern, itin>, T8PD,
VEX_4V, FMASC, Requires<[HasFMA, NoAVX512]>;
// FMA4 Instruction Templates
class FMA4<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag>pattern, InstrItinClass itin = NoItinerary>
: Ii8Reg<o, F, outs, ins, asm, pattern, itin>, TAPD,
VEX_4V, FMASC, Requires<[HasFMA4]>;
VEX_4V, FMASC, Requires<[HasFMA4, NoVLX]>;
class FMA4S<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag>pattern, InstrItinClass itin = NoItinerary>
: Ii8Reg<o, F, outs, ins, asm, pattern, itin>, TAPD,
VEX_4V, FMASC, Requires<[HasFMA4, NoAVX512]>;
// XOP 2, 3 and 4 Operand Instruction Template
class IXOP<bits<8> o, Format F, dag outs, dag ins, string asm,