diff options
Diffstat (limited to 'llvm/lib/Target')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrAVX512.td | 39 |
1 files changed, 24 insertions, 15 deletions
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index 75a265bf8a1..d44d1395f24 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -1588,21 +1588,6 @@ defm VPBLENDMB : blendmask_bw <0x66, "vpblendmb", avx512vl_i8_info>; defm VPBLENDMW : blendmask_bw <0x66, "vpblendmw", avx512vl_i16_info>, VEX_W; -let Predicates = [HasAVX512, NoVLX] in { -def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1), - (v8f32 VR256X:$src2))), - (EXTRACT_SUBREG - (v16f32 (VBLENDMPSZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), - (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)), - (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), sub_ymm)>; - -def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1), - (v8i32 VR256X:$src2))), - (EXTRACT_SUBREG - (v16i32 (VPBLENDMDZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), - (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)), - (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), sub_ymm)>; -} //===----------------------------------------------------------------------===// // Compare Instructions //===----------------------------------------------------------------------===// @@ -2978,6 +2963,30 @@ def : Pat<(v16i32 (vselect (xor VK16:$mask, (v16i1 immAllOnesV)), (v16i32 VR512:$src))), (VMOVDQA32Zrrkz VK16WM:$mask, VR512:$src)>; +// Patterns for handling v8i1 selects of 256-bit vectors when VLX isn't +// available. Use a 512-bit operation and extract. +let Predicates = [HasAVX512, NoVLX] in { +def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1), + (v8f32 VR256X:$src0))), + (EXTRACT_SUBREG + (v16f32 + (VMOVAPSZrrk + (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src0, sub_ymm)), + (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), + (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), + sub_ymm)>; + +def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1), + (v8i32 VR256X:$src0))), + (EXTRACT_SUBREG + (v16i32 + (VMOVDQA32Zrrk + (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src0, sub_ymm)), + (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), + (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), + sub_ymm)>; +} + let Predicates = [HasVLX, NoBWI] in { // 128-bit load/store without BWI. def : Pat<(alignedstore (v8i16 VR128X:$src), addr:$dst), |
