1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-20 19:42:54 +02:00

Add AVX SSE4.2 instructions

llvm-svn: 107752
This commit is contained in:
Bruno Cardoso Lopes 2010-07-07 03:39:29 +00:00
parent 931471d7e8
commit 6222076cd1
3 changed files with 258 additions and 114 deletions

View File

@ -4705,36 +4705,196 @@ def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
OpSize;
//===----------------------------------------------------------------------===//
// SSE4.2 Instructions
// SSE4.2 - Compare Instructions
//===----------------------------------------------------------------------===//
/// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
let Constraints = "$src1 = $dst" in {
multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
Intrinsic IntId128, bit Commutable = 0> {
def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
OpSize {
let isCommutable = Commutable;
}
def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst,
(IntId128 VR128:$src1,
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
Intrinsic IntId128, bit Is2Addr = 1> {
def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
OpSize;
def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
(IntId128 VR128:$src1,
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in
defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
(PCMPGTQrr VR128:$src1, VR128:$src2)>;
def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
(PCMPGTQrm VR128:$src1, addr:$src2)>;
//===----------------------------------------------------------------------===//
// SSE4.2 - String/text Processing Instructions
//===----------------------------------------------------------------------===//
// Packed Compare Implicit Length Strings, Return Mask
let Defs = [EFLAGS], usesCustomInserter = 1 in {
def PCMPISTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"#PCMPISTRM128rr PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
imm:$src3))]>, OpSize;
def PCMPISTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"#PCMPISTRM128rm PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128
VR128:$src1, (load addr:$src2), imm:$src3))]>, OpSize;
}
let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
Predicates = [HasAVX, HasSSE42] in {
def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
}
let Defs = [XMM0, EFLAGS] in {
def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
}
// Packed Compare Explicit Length Strings, Return Mask
let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"#PCMPESTRM128rr PSEUDO!",
[(set VR128:$dst,
(int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>, OpSize;
def PCMPESTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"#PCMPESTRM128rm PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
OpSize;
}
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42],
Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
}
let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
}
// Packed Compare Implicit Length Strings, Return Index
let Defs = [ECX, EFLAGS] in {
multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
def rr : SS42AI<0x63, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
[(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
(implicit EFLAGS)]>, OpSize;
def rm : SS42AI<0x63, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
[(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
(implicit EFLAGS)]>, OpSize;
}
}
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in {
defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
VEX;
defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
VEX;
defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
VEX;
defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
VEX;
defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
VEX;
defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
VEX;
}
defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
// Packed Compare Explicit Length Strings, Return Index
let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
def rr : SS42AI<0x61, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
[(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
(implicit EFLAGS)]>, OpSize;
def rm : SS42AI<0x61, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
[(set ECX,
(IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
(implicit EFLAGS)]>, OpSize;
}
}
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in {
defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
VEX;
defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
VEX;
defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
VEX;
defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
VEX;
defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
VEX;
defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
VEX;
}
defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
//===----------------------------------------------------------------------===//
// SSE4.2 - CRC Instructions
//===----------------------------------------------------------------------===//
// No CRC instructions have AVX equivalents
// crc intrinsic instruction
// This set of instructions are only rm, the only difference is the size
// of r and m.
@ -4802,101 +4962,6 @@ let Constraints = "$src1 = $dst" in {
REX_W;
}
// String/text processing instructions.
let Defs = [EFLAGS], usesCustomInserter = 1 in {
def PCMPISTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"#PCMPISTRM128rr PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
imm:$src3))]>, OpSize;
def PCMPISTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"#PCMPISTRM128rm PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, (load addr:$src2),
imm:$src3))]>, OpSize;
}
let Defs = [XMM0, EFLAGS] in {
def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
}
let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"#PCMPESTRM128rr PSEUDO!",
[(set VR128:$dst,
(int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>, OpSize;
def PCMPESTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"#PCMPESTRM128rm PSEUDO!",
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
OpSize;
}
let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
}
let Defs = [ECX, EFLAGS] in {
multiclass SS42AI_pcmpistri<Intrinsic IntId128> {
def rr : SS42AI<0x63, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"pcmpistri\t{$src3, $src2, $src1|$src1, $src2, $src3}",
[(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
(implicit EFLAGS)]>, OpSize;
def rm : SS42AI<0x63, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"pcmpistri\t{$src3, $src2, $src1|$src1, $src2, $src3}",
[(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
(implicit EFLAGS)]>, OpSize;
}
}
defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
let Defs = [ECX, EFLAGS] in {
let Uses = [EAX, EDX] in {
multiclass SS42AI_pcmpestri<Intrinsic IntId128> {
def rr : SS42AI<0x61, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"pcmpestri\t{$src5, $src3, $src1|$src1, $src3, $src5}",
[(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
(implicit EFLAGS)]>, OpSize;
def rm : SS42AI<0x61, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"pcmpestri\t{$src5, $src3, $src1|$src1, $src3, $src5}",
[(set ECX,
(IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
(implicit EFLAGS)]>, OpSize;
}
}
}
defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
//===----------------------------------------------------------------------===//
// AES-NI Instructions
//===----------------------------------------------------------------------===//

View File

@ -12122,3 +12122,43 @@
// CHECK: encoding: [0xc4,0xe2,0x79,0x2a,0x10]
vmovntdqa (%eax), %xmm2
// CHECK: vpcmpgtq %xmm2, %xmm5, %xmm1
// CHECK: encoding: [0xc4,0xe2,0x51,0x37,0xca]
vpcmpgtq %xmm2, %xmm5, %xmm1
// CHECK: vpcmpgtq (%eax), %xmm5, %xmm3
// CHECK: encoding: [0xc4,0xe2,0x51,0x37,0x18]
vpcmpgtq (%eax), %xmm5, %xmm3
// CHECK: vpcmpistrm $7, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x62,0xea,0x07]
vpcmpistrm $7, %xmm2, %xmm5
// CHECK: vpcmpistrm $7, (%eax), %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x62,0x28,0x07]
vpcmpistrm $7, (%eax), %xmm5
// CHECK: vpcmpestrm $7, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x60,0xea,0x07]
vpcmpestrm $7, %xmm2, %xmm5
// CHECK: vpcmpestrm $7, (%eax), %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x60,0x28,0x07]
vpcmpestrm $7, (%eax), %xmm5
// CHECK: vpcmpistri $7, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x63,0xea,0x07]
vpcmpistri $7, %xmm2, %xmm5
// CHECK: vpcmpistri $7, (%eax), %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x63,0x28,0x07]
vpcmpistri $7, (%eax), %xmm5
// CHECK: vpcmpestri $7, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x61,0xea,0x07]
vpcmpestri $7, %xmm2, %xmm5
// CHECK: vpcmpestri $7, (%eax), %xmm5
// CHECK: encoding: [0xc4,0xe3,0x79,0x61,0x28,0x07]
vpcmpestri $7, (%eax), %xmm5

View File

@ -2186,3 +2186,42 @@ pshufb CPI1_0(%rip), %xmm1
// CHECK: encoding: [0xc4,0x62,0x79,0x2a,0x20]
vmovntdqa (%rax), %xmm12
// CHECK: vpcmpgtq %xmm12, %xmm10, %xmm11
// CHECK: encoding: [0xc4,0x42,0x29,0x37,0xdc]
vpcmpgtq %xmm12, %xmm10, %xmm11
// CHECK: vpcmpgtq (%rax), %xmm10, %xmm13
// CHECK: encoding: [0xc4,0x62,0x29,0x37,0x28]
vpcmpgtq (%rax), %xmm10, %xmm13
// CHECK: vpcmpistrm $7, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x43,0x79,0x62,0xd4,0x07]
vpcmpistrm $7, %xmm12, %xmm10
// CHECK: vpcmpistrm $7, (%rax), %xmm10
// CHECK: encoding: [0xc4,0x63,0x79,0x62,0x10,0x07]
vpcmpistrm $7, (%rax), %xmm10
// CHECK: vpcmpestrm $7, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x43,0x79,0x60,0xd4,0x07]
vpcmpestrm $7, %xmm12, %xmm10
// CHECK: vpcmpestrm $7, (%rax), %xmm10
// CHECK: encoding: [0xc4,0x63,0x79,0x60,0x10,0x07]
vpcmpestrm $7, (%rax), %xmm10
// CHECK: vpcmpistri $7, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x43,0x79,0x63,0xd4,0x07]
vpcmpistri $7, %xmm12, %xmm10
// CHECK: vpcmpistri $7, (%rax), %xmm10
// CHECK: encoding: [0xc4,0x63,0x79,0x63,0x10,0x07]
vpcmpistri $7, (%rax), %xmm10
// CHECK: vpcmpestri $7, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x43,0x79,0x61,0xd4,0x07]
vpcmpestri $7, %xmm12, %xmm10
// CHECK: vpcmpestri $7, (%rax), %xmm10
// CHECK: encoding: [0xc4,0x63,0x79,0x61,0x10,0x07]
vpcmpestri $7, (%rax), %xmm10