RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
Code Issues Projects Releases Wiki Activity

[AVX512] Change the AddedComplexity on some patterns to match their AVX/SSE equivalents. This helps group them close together in the isel tables and enable table compression.

Browse Source 
llvm-svn: 270354
This commit is contained in:
Craig Topper 2016-05-22 06:09:34 +00:00
parent b77f901eb5
commit b6b424b279
1 changed files with 13 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
lib/Target/X86/X86InstrAVX512.td
View File

@ -3138,14 +3138,21 @@ def VMOVZPQILo2PQIZrm : AVX512XSI<0x7E, MRMSrcMem, (outs VR128X:$dst),
EVEX_CD8<8, CD8VT8>;
let Predicates = [HasAVX512] in {
let AddedComplexity = 15 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
(VMOVDI2PDIZrr GR32:$src)>;
def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
(VMOV64toPQIZrr GR64:$src)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
(SUBREG_TO_REG (i64 0), (VMOV64toPQIZrr GR64:$src), sub_xmm)>;
}
// AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
let AddedComplexity = 20 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
(VMOVDI2PDIZrm addr:$src)>;
def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
(VMOV64toPQIZrr GR64:$src)>;
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
(VMOVDI2PDIZrr GR32:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
(VMOVDI2PDIZrm addr:$src)>;
@ -3157,17 +3164,14 @@ let Predicates = [HasAVX512] in {
(VMOVZPQILo2PQIZrr VR128X:$src)>;
def : Pat<(v2i64 (X86vzload addr:$src)),
(VMOVZPQILo2PQIZrm addr:$src)>;
def : Pat<(v4i64 (X86vzload addr:$src)),
(SUBREG_TO_REG (i64 0), (VMOVZPQILo2PQIZrm addr:$src), sub_xmm)>;
}
// Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
(v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
(SUBREG_TO_REG (i32 0), (VMOVDI2PDIZrr GR32:$src), sub_xmm)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
(SUBREG_TO_REG (i64 0), (VMOV64toPQIZrr GR64:$src), sub_xmm)>;
def : Pat<(v4i64 (X86vzload addr:$src)),
(SUBREG_TO_REG (i64 0), (VMOVZPQILo2PQIZrm addr:$src), sub_xmm)>;
// Use regular 128-bit instructions to match 512-bit scalar_to_vec+zext.
def : Pat<(v8i64 (X86vzload addr:$src)),