RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-12-04 01:11:44 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[X86][AVX2] Passing the appropriate memory operand class to VPMADDWD instruction.

Browse Source 
Replacing the memory operand in the ymm version of VPMADDWD from i128mem to i256mem.

Differential Revision: https://reviews.llvm.org/D28024

llvm-svn: 290333
This commit is contained in:
Ayman Musa 2016-12-22 08:42:46 +00:00
parent 8c6f59de4c
commit 6f11282fa7
3 changed files with 115 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Target/X86/X86InstrSSE.td
View File

@ -3952,7 +3952,7 @@ defm VPMADDWD : PDI_binop_rm2<0xF5, "vpmaddwd", X86vpmaddwd, v4i32, v8i16, VR128
let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in
defm VPMADDWDY : PDI_binop_rm2<0xF5, "vpmaddwd", X86vpmaddwd, v8i32, v16i16,
VR256, loadv4i64, i128mem, SSE_PMADD,
VR256, loadv4i64, i256mem, SSE_PMADD,
0>, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in
defm PMADDWD : PDI_binop_rm2<0xF5, "pmaddwd", X86vpmaddwd, v4i32, v8i16, VR128,

56
test/MC/X86/intel-syntax-x86-64-avx.s
View File

@ -159,3 +159,59 @@
// CHECK: vcvttpd2dq xmm0, ymmword ptr [rax]
// CHECK: encoding: [0xc5,0xfd,0xe6,0x00]
vcvttpd2dqy xmm0, ymmword ptr [rax]
// CHECK: vpmaddwd xmm1, xmm2, xmm3
// CHECK: encoding: [0xc5,0xe9,0xf5,0xcb]
vpmaddwd xmm1, xmm2, xmm3
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rcx]
// CHECK: encoding: [0xc5,0xe9,0xf5,0x09]
vpmaddwd xmm1, xmm2, xmmword ptr [rcx]
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rsp - 4]
// CHECK: encoding: [0xc5,0xe9,0xf5,0x4c,0x24,0xfc]
vpmaddwd xmm1, xmm2, xmmword ptr [rsp - 4]
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rsp + 4]
// CHECK: encoding: [0xc5,0xe9,0xf5,0x4c,0x24,0x04]
vpmaddwd xmm1, xmm2, xmmword ptr [rsp + 4]
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 + 268435456]
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x00,0x00,0x00,0x10]
vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 + 268435456]
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 - 536870912]
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x00,0x00,0x00,0xe0]
vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 - 536870912]
// CHECK: vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 - 536870910]
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x02,0x00,0x00,0xe0]
vpmaddwd xmm1, xmm2, xmmword ptr [rcx + 8*r14 - 536870910]
// CHECK: vpmaddwd ymm1, ymm2, ymm3
// CHECK: encoding: [0xc5,0xed,0xf5,0xcb]
vpmaddwd ymm1, ymm2, ymm3
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rcx]
// CHECK: encoding: [0xc5,0xed,0xf5,0x09]
vpmaddwd ymm1, ymm2, ymmword ptr [rcx]
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rsp - 4]
// CHECK: encoding: [0xc5,0xed,0xf5,0x4c,0x24,0xfc]
vpmaddwd ymm1, ymm2, ymmword ptr [rsp - 4]
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rsp + 4]
// CHECK: encoding: [0xc5,0xed,0xf5,0x4c,0x24,0x04]
vpmaddwd ymm1, ymm2, ymmword ptr [rsp + 4]
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 + 268435456]
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x00,0x00,0x00,0x10]
vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 + 268435456]
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 - 536870912]
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x00,0x00,0x00,0xe0]
vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 - 536870912]
// CHECK: vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 - 536870910]
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x02,0x00,0x00,0xe0]
vpmaddwd ymm1, ymm2, ymmword ptr [rcx + 8*r14 - 536870910]

58
test/MC/X86/x86_64-avx-encoding.s
View File

@ -4245,3 +4245,61 @@ _foo2:
// CHECK: vmovsd %xmm8, %xmm0, %xmm0
// CHECK: encoding: [0xc5,0x7b,0x11,0xc0]
vmovsd %xmm8, %xmm0, %xmm0
// CHECK: vpmaddwd %xmm3, %xmm2, %xmm1
// CHECK: encoding: [0xc5,0xe9,0xf5,0xcb]
vpmaddwd %xmm3, %xmm2, %xmm1
// CHECK: vpmaddwd (%rcx), %xmm2, %xmm1
// CHECK: encoding: [0xc5,0xe9,0xf5,0x09]
vpmaddwd (%rcx), %xmm2, %xmm1
// CHECK: vpmaddwd -4(%rsp), %xmm2, %xmm1
// CHECK: encoding: [0xc5,0xe9,0xf5,0x4c,0x24,0xfc]
vpmaddwd -4(%rsp), %xmm2, %xmm1
// CHECK: vpmaddwd 4(%rsp), %xmm2, %xmm1
// CHECK: encoding: [0xc5,0xe9,0xf5,0x4c,0x24,0x04]
vpmaddwd 4(%rsp), %xmm2, %xmm1
// CHECK: vpmaddwd 268435456(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x00,0x00,0x00,0x10]
vpmaddwd 268435456(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: vpmaddwd -536870912(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x00,0x00,0x00,0xe0]
vpmaddwd -536870912(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: vpmaddwd -536870910(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: encoding: [0xc4,0xa1,0x69,0xf5,0x8c,0xf1,0x02,0x00,0x00,0xe0]
vpmaddwd -536870910(%rcx,%r14,8), %xmm2, %xmm1
// CHECK: vpmaddwd %ymm3, %ymm2, %ymm1
// CHECK: encoding: [0xc5,0xed,0xf5,0xcb]
vpmaddwd %ymm3, %ymm2, %ymm1
// CHECK: vpmaddwd (%rcx), %ymm2, %ymm1
// CHECK: encoding: [0xc5,0xed,0xf5,0x09]
vpmaddwd (%rcx), %ymm2, %ymm1
// CHECK: vpmaddwd -4(%rsp), %ymm2, %ymm1
// CHECK: encoding: [0xc5,0xed,0xf5,0x4c,0x24,0xfc]
vpmaddwd -4(%rsp), %ymm2, %ymm1
// CHECK: vpmaddwd 4(%rsp), %ymm2, %ymm1
// CHECK: encoding: [0xc5,0xed,0xf5,0x4c,0x24,0x04]
vpmaddwd 4(%rsp), %ymm2, %ymm1
// CHECK: vpmaddwd 268435456(%rcx,%r14,8), %ymm2, %ymm1
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x00,0x00,0x00,0x10]
vpmaddwd 268435456(%rcx,%r14,8), %ymm2, %ymm1
// CHECK: vpmaddwd -536870912(%rcx,%r14,8), %ymm2, %ymm1
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x00,0x00,0x00,0xe0]
vpmaddwd -536870912(%rcx,%r14,8), %ymm2, %ymm1
// CHECK: vpmaddwd -536870910(%rcx,%r14,8), %ymm2, %ymm1
// CHECK: encoding: [0xc4,0xa1,0x6d,0xf5,0x8c,0xf1,0x02,0x00,0x00,0xe0]
vpmaddwd -536870910(%rcx,%r14,8), %ymm2, %ymm1