RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-20 19:42:54 +02:00
Code Issues Projects Releases Wiki Activity

Add AVX SSE4.1 extractps and pinsr instructions

Browse Source 
llvm-svn: 107746
This commit is contained in:
Bruno Cardoso Lopes 2010-07-07 01:01:13 +00:00
parent 3f4e5779d7
commit fa10461265
3 changed files with 139 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

102
lib/Target/X86/X86InstrSSE.td
View File

@ -4516,6 +4516,8 @@ multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
addr:$dst)]>, OpSize;
}
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
// Also match an EXTRACTPS store when the store is done as f32 instead of i32.
@ -4525,46 +4527,76 @@ def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
(EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
Requires<[HasSSE41]>;
let Constraints = "$src1 = $dst" in {
multiclass SS41I_insert8<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
imm:$src3))]>, OpSize;
}
multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
imm:$src3))]>, OpSize;
}
defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
let Constraints = "$src1 = $dst" in {
multiclass SS41I_insert32<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
imm:$src3)))]>, OpSize;
}
multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
imm:$src3)))]>, OpSize;
}
defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
multiclass SS41I_insert64_avx<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst,
(v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
OpSize, REX_W;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst,
(v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
imm:$src3)))]>, OpSize, REX_W;
}
let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
defm VPINSRQ : SS41I_insert64_avx<0x22, "vpinsrq">, VEX_4V, VEX_W;
// insertps has a few different modes, there's the first two here below which
// are optimized inserts that won't zero arbitrary elements in the destination

32
test/MC/AsmParser/X86/x86_32-encoding.s
View File

@ -12070,3 +12070,35 @@
// CHECK: encoding: [0xc4,0xe3,0x79,0x14,0x10,0x07]
vpextrb $7, %xmm2, (%eax)
// CHECK: vextractps $7, %xmm2, (%eax)
// CHECK: encoding: [0xc4,0xe3,0x79,0x17,0x10,0x07]
vextractps $7, %xmm2, (%eax)
// CHECK: vextractps $7, %xmm2, %eax
// CHECK: encoding: [0xc4,0xe3,0x79,0x17,0xd0,0x07]
vextractps $7, %xmm2, %eax
// CHECK: vpinsrw $7, %eax, %xmm2, %xmm5
// CHECK: encoding: [0xc5,0xe9,0xc4,0xe8,0x07]
vpinsrw $7, %eax, %xmm2, %xmm5
// CHECK: vpinsrw $7, (%eax), %xmm2, %xmm5
// CHECK: encoding: [0xc5,0xe9,0xc4,0x28,0x07]
vpinsrw $7, (%eax), %xmm2, %xmm5
// CHECK: vpinsrb $7, %eax, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x69,0x20,0xe8,0x07]
vpinsrb $7, %eax, %xmm2, %xmm5
// CHECK: vpinsrb $7, (%eax), %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x69,0x20,0x28,0x07]
vpinsrb $7, (%eax), %xmm2, %xmm5
// CHECK: vpinsrd $7, %eax, %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x69,0x22,0xe8,0x07]
vpinsrd $7, %eax, %xmm2, %xmm5
// CHECK: vpinsrd $7, (%eax), %xmm2, %xmm5
// CHECK: encoding: [0xc4,0xe3,0x69,0x22,0x28,0x07]
vpinsrd $7, (%eax), %xmm2, %xmm5

40
test/MC/AsmParser/X86/x86_64-encoding.s
View File

@ -2126,3 +2126,43 @@ pshufb CPI1_0(%rip), %xmm1
// CHECK: encoding: [0xc4,0x63,0xf9,0x16,0x21,0x07]
vpextrq $7, %xmm12, (%rcx)
// CHECK: vextractps $7, %xmm12, (%rax)
// CHECK: encoding: [0xc4,0x63,0x79,0x17,0x20,0x07]
vextractps $7, %xmm12, (%rax)
// CHECK: vextractps $7, %xmm12, %eax
// CHECK: encoding: [0xc4,0x63,0x79,0x17,0xe0,0x07]
vextractps $7, %xmm12, %eax
// CHECK: vpinsrw $7, %eax, %xmm12, %xmm10
// CHECK: encoding: [0xc5,0x19,0xc4,0xd0,0x07]
vpinsrw $7, %eax, %xmm12, %xmm10
// CHECK: vpinsrw $7, (%rax), %xmm12, %xmm10
// CHECK: encoding: [0xc5,0x19,0xc4,0x10,0x07]
vpinsrw $7, (%rax), %xmm12, %xmm10
// CHECK: vpinsrb $7, %eax, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x19,0x20,0xd0,0x07]
vpinsrb $7, %eax, %xmm12, %xmm10
// CHECK: vpinsrb $7, (%rax), %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x19,0x20,0x10,0x07]
vpinsrb $7, (%rax), %xmm12, %xmm10
// CHECK: vpinsrd $7, %eax, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x19,0x22,0xd0,0x07]
vpinsrd $7, %eax, %xmm12, %xmm10
// CHECK: vpinsrd $7, (%rax), %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x19,0x22,0x10,0x07]
vpinsrd $7, (%rax), %xmm12, %xmm10
// CHECK: vpinsrq $7, %rax, %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x99,0x22,0xd0,0x07]
vpinsrq $7, %rax, %xmm12, %xmm10
// CHECK: vpinsrq $7, (%rax), %xmm12, %xmm10
// CHECK: encoding: [0xc4,0x63,0x99,0x22,0x10,0x07]
vpinsrq $7, (%rax), %xmm12, %xmm10