[PATCH] D128669: [SVE] Use CPY to zero active lanes of a floating point vector.

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm created this revision.
Herald added subscribers: psnobl, hiraditya, tschuett.
Herald added a reviewer: efriedma.
Herald added a project: All.
paulwalker-arm requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Patterns exist for the integer case that are trivially expandable
to cover 0.0f.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D128669

Files:
  llvm/lib/Target/AArch64/SVEInstrFormats.td
  llvm/test/CodeGen/AArch64/sve-vselect-imm.ll


Index: llvm/test/CodeGen/AArch64/sve-vselect-imm.ll
===================================================================
--- llvm/test/CodeGen/AArch64/sve-vselect-imm.ll
+++ llvm/test/CodeGen/AArch64/sve-vselect-imm.ll
@@ -311,8 +311,7 @@
 define <vscale x 8 x half> @sel_merge_nxv8f16_zero(<vscale x 8 x i1> %p, <vscale x 8 x half> %in) {
 ; CHECK-LABEL: sel_merge_nxv8f16_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.h, #0 // =0x0
-; CHECK-NEXT:    mov z0.h, p0/m, z1.h
+; CHECK-NEXT:    mov z0.h, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 8 x i1> %p, <vscale x 8 x half> zeroinitializer, <vscale x 8 x half> %in
 ret <vscale x 8 x half> %sel
@@ -321,8 +320,7 @@
 define <vscale x 4 x half> @sel_merge_nx4f16_zero(<vscale x 4 x i1> %p, <vscale x 4 x half> %in) {
 ; CHECK-LABEL: sel_merge_nx4f16_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.h, #0 // =0x0
-; CHECK-NEXT:    mov z0.s, p0/m, z1.s
+; CHECK-NEXT:    mov z0.s, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 4 x i1> %p, <vscale x 4 x half> zeroinitializer, <vscale x 4 x half> %in
 ret <vscale x 4 x half> %sel
@@ -331,8 +329,7 @@
 define <vscale x 2 x half> @sel_merge_nx2f16_zero(<vscale x 2 x i1> %p, <vscale x 2 x half> %in) {
 ; CHECK-LABEL: sel_merge_nx2f16_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.h, #0 // =0x0
-; CHECK-NEXT:    mov z0.d, p0/m, z1.d
+; CHECK-NEXT:    mov z0.d, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 2 x i1> %p, <vscale x 2 x half> zeroinitializer, <vscale x 2 x half> %in
 ret <vscale x 2 x half> %sel
@@ -341,8 +338,7 @@
 define <vscale x 4 x float> @sel_merge_nx4f32_zero(<vscale x 4 x i1> %p, <vscale x 4 x float> %in) {
 ; CHECK-LABEL: sel_merge_nx4f32_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.s, #0 // =0x0
-; CHECK-NEXT:    mov z0.s, p0/m, z1.s
+; CHECK-NEXT:    mov z0.s, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 4 x i1> %p, <vscale x 4 x float> zeroinitializer, <vscale x 4 x float> %in
 ret <vscale x 4 x float> %sel
@@ -351,8 +347,7 @@
 define <vscale x 2 x float> @sel_merge_nx2f32_zero(<vscale x 2 x i1> %p, <vscale x 2 x float> %in) {
 ; CHECK-LABEL: sel_merge_nx2f32_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.s, #0 // =0x0
-; CHECK-NEXT:    mov z0.d, p0/m, z1.d
+; CHECK-NEXT:    mov z0.d, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 2 x i1> %p, <vscale x 2 x float> zeroinitializer, <vscale x 2 x float> %in
 ret <vscale x 2 x float> %sel
@@ -361,8 +356,7 @@
 define <vscale x 2 x double> @sel_merge_nx2f64_zero(<vscale x 2 x i1> %p, <vscale x 2 x double> %in) {
 ; CHECK-LABEL: sel_merge_nx2f64_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z1.d, #0 // =0x0
-; CHECK-NEXT:    mov z0.d, p0/m, z1.d
+; CHECK-NEXT:    mov z0.d, p0/m, #0 // =0x0
 ; CHECK-NEXT:    ret
 %sel = select <vscale x 2 x i1> %p, <vscale x 2 x double> zeroinitializer, <vscale x 2 x double> %in
 ret <vscale x 2 x double> %sel
Index: llvm/lib/Target/AArch64/SVEInstrFormats.td
===================================================================
--- llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -4648,6 +4648,19 @@
                   (!cast<Instruction>(NAME # _S) ZPR32:$Zd, PPRAny:$Pg, 0, 0), 0>;
   def : InstAlias<"fmov $Zd, $Pg/m, #0.0",
                   (!cast<Instruction>(NAME # _D) ZPR64:$Zd, PPRAny:$Pg, 0, 0), 0>;
+
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv8f16 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _H) $Zd, $Pg, 0, 0)>;
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv4f16 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _S) $Zd, $Pg, 0, 0)>;
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv2f16 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _D) $Zd, $Pg, 0, 0)>;
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv4f32 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _S) $Zd, $Pg, 0, 0)>;
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv2f32 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _D) $Zd, $Pg, 0, 0)>;
+  def : Pat<(vselect PPRAny:$Pg, (SVEDup0), (nxv2f64 ZPR:$Zd)),
+            (!cast<Instruction>(NAME # _D) $Zd, $Pg, 0, 0)>;
 }
 
 multiclass sve_int_dup_imm_pred_zero_inst<


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D128669.440299.patch
Type: text/x-patch
Size: 4248 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220627/2055c927/attachment.bin>