[llvm] r312694 - [X86] Remove patterns for selecting a v8f32 X86ISD::MOVSS or v4f64 X86ISD::MOVSD.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 6 22:08:17 PDT 2017
Author: ctopper
Date: Wed Sep 6 22:08:16 2017
New Revision: 312694
URL: http://llvm.org/viewvc/llvm-project?rev=312694&view=rev
Log:
[X86] Remove patterns for selecting a v8f32 X86ISD::MOVSS or v4f64 X86ISD::MOVSD.
I don't think we ever generate these. If we did, I would expect we would also be able to generate v16f32 and v8f64, but we don't have those patterns.
Modified:
llvm/trunk/lib/Target/X86/X86InstrAVX512.td
llvm/trunk/lib/Target/X86/X86InstrSSE.td
Modified: llvm/trunk/lib/Target/X86/X86InstrAVX512.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrAVX512.td?rev=312694&r1=312693&r2=312694&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrAVX512.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrAVX512.td Wed Sep 6 22:08:16 2017
@@ -4283,36 +4283,12 @@ let Predicates = [HasAVX512] in {
(VMOVSSZrr (v4f32 VR128X:$src1),
(COPY_TO_REGCLASS (v4f32 VR128X:$src2), FR32X))>;
- // 256-bit variants
- def : Pat<(v8i32 (X86Movss VR256X:$src1, VR256X:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSZrr (EXTRACT_SUBREG (v8i32 VR256X:$src1), sub_xmm),
- (EXTRACT_SUBREG (v8i32 VR256X:$src2), sub_xmm)),
- sub_xmm)>;
- def : Pat<(v8f32 (X86Movss VR256X:$src1, VR256X:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSZrr (EXTRACT_SUBREG (v8f32 VR256X:$src1), sub_xmm),
- (EXTRACT_SUBREG (v8f32 VR256X:$src2), sub_xmm)),
- sub_xmm)>;
-
// Shuffle with VMOVSD
def : Pat<(v2i64 (X86Movsd VR128X:$src1, VR128X:$src2)),
(VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
def : Pat<(v2f64 (X86Movsd VR128X:$src1, VR128X:$src2)),
(VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
- // 256-bit variants
- def : Pat<(v4i64 (X86Movsd VR256X:$src1, VR256X:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDZrr (EXTRACT_SUBREG (v4i64 VR256X:$src1), sub_xmm),
- (EXTRACT_SUBREG (v4i64 VR256X:$src2), sub_xmm)),
- sub_xmm)>;
- def : Pat<(v4f64 (X86Movsd VR256X:$src1, VR256X:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDZrr (EXTRACT_SUBREG (v4f64 VR256X:$src1), sub_xmm),
- (EXTRACT_SUBREG (v4f64 VR256X:$src2), sub_xmm)),
- sub_xmm)>;
-
def : Pat<(v2f64 (X86Movlpd VR128X:$src1, VR128X:$src2)),
(VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
def : Pat<(v2i64 (X86Movlpd VR128X:$src1, VR128X:$src2)),
Modified: llvm/trunk/lib/Target/X86/X86InstrSSE.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrSSE.td?rev=312694&r1=312693&r2=312694&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrSSE.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrSSE.td Wed Sep 6 22:08:16 2017
@@ -512,36 +512,12 @@ let Predicates = [UseAVX] in {
(VMOVSSrr (v4f32 VR128:$src1),
(COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
- // 256-bit variants
- def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
- (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
- sub_xmm)>;
- def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
- (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
- sub_xmm)>;
-
// Shuffle with VMOVSD
def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
(VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
(VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
- // 256-bit variants
- def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
- (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
- sub_xmm)>;
- def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
- (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
- sub_xmm)>;
-
// FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
// is during lowering, where it's not possible to recognize the fold cause
// it has two uses through a bitcast. One use disappears at isel time and the
More information about the llvm-commits
mailing list