[llvm] be26e6f - [AArch64][SVE] Remove redundant PTEST following PNEXT/PFIRST

[Peter Waller via llvm-commits]

Author: Peter Waller
Date: 2021-10-05T15:10:48Z
New Revision: be26e6ff737d89827f0eb5678cbc88aae2068348

URL: https://github.com/llvm/llvm-project/commit/be26e6ff737d89827f0eb5678cbc88aae2068348
DIFF: https://github.com/llvm/llvm-project/commit/be26e6ff737d89827f0eb5678cbc88aae2068348.diff

LOG: [AArch64][SVE] Remove redundant PTEST following PNEXT/PFIRST

PNEXT and PFIRST set the NZCV flags, so the subsequent PTEST can be
optimized away in AArch64InstrInfo::optimizePTestInstr.

See-also: https://reviews.llvm.org/D93292

Differential Revision: https://reviews.llvm.org/D110177

Added: 
    llvm/test/CodeGen/AArch64/sve-ptest-removal-pfirst-pnext.ll

Modified: 
    llvm/lib/Target/AArch64/SVEInstrFormats.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index e9e4098b6ee6..a2a0d2b0dae6 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -627,6 +627,8 @@ class sve_int_pfirst_next<bits<2> sz8_64, bits<5> opc, string asm,
 
   let Constraints = "$Pdn = $_Pdn";
   let Defs = [NZCV];
+  let isPTestLike = 1;
+  let ElementSize = pprty.ElementSize;
 }
 
 multiclass sve_int_pfirst<bits<5> opc, string asm, SDPatternOperator op> {

diff  --git a/llvm/test/CodeGen/AArch64/sve-ptest-removal-pfirst-pnext.ll b/llvm/test/CodeGen/AArch64/sve-ptest-removal-pfirst-pnext.ll
new file mode 100644
index 000000000000..55af332eacf9
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-ptest-removal-pfirst-pnext.ll
@@ -0,0 +1,74 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve %s -o - | FileCheck %s
+
+define i32 @pfirst_16(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
+; CHECK-LABEL: pfirst_16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    pfirst p1.b, p0, p1.b
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pfirst.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+define i32 @pnext_2(<vscale x 2 x i1> %pg, <vscale x 2 x i1> %a) {
+; CHECK-LABEL: pnext_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    pnext p1.d, p0, p1.d
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.pnext.nxv2i1(<vscale x 2 x i1> %pg, <vscale x 2 x i1> %a)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv2i1(<vscale x 2 x i1> %pg, <vscale x 2 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+define i32 @pnext_4(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %a) {
+; CHECK-LABEL: pnext_4:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    pnext p1.s, p0, p1.s
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.pnext.nxv4i1(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %a)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv4i1(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+define i32 @pnext_8(<vscale x 8 x i1> %pg, <vscale x 8 x i1> %a) {
+; CHECK-LABEL: pnext_8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    pnext p1.h, p0, p1.h
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.pnext.nxv8i1(<vscale x 8 x i1> %pg, <vscale x 8 x i1> %a)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv8i1(<vscale x 8 x i1> %pg, <vscale x 8 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+define i32 @pnext_16(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
+; CHECK-LABEL: pnext_16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    pnext p1.b, p0, p1.b
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pnext.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+declare <vscale x 16 x i1> @llvm.aarch64.sve.pfirst.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
+
+declare <vscale x 16 x i1> @llvm.aarch64.sve.pnext.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
+declare <vscale x 8 x i1> @llvm.aarch64.sve.pnext.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.pnext.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.pnext.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)
+
+declare i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
+declare i1 @llvm.aarch64.sve.ptest.any.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>)
+declare i1 @llvm.aarch64.sve.ptest.any.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>)
+declare i1 @llvm.aarch64.sve.ptest.any.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)