[llvm] f1ade1f - [LLVM][CodeGen][AArch64] while_le(#,max_int) -> all_active (#111183)

[via llvm-commits]

Author: Paul Walker
Date: 2024-10-22T12:28:39+01:00
New Revision: f1ade1f874db066a46142cacbb67f80d272862ed

URL: https://github.com/llvm/llvm-project/commit/f1ade1f874db066a46142cacbb67f80d272862ed
DIFF: https://github.com/llvm/llvm-project/commit/f1ade1f874db066a46142cacbb67f80d272862ed.diff

LOG: [LLVM][CodeGen][AArch64] while_le(#,max_int) -> all_active (#111183)

When the second operand of an incrementing while instruction is the
maximum value, comparisons that include equality can never fail.

Added: 
    

Modified: 
    llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
    llvm/test/CodeGen/AArch64/sve-intrinsics-while.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 018d48b2c8f12e..927c057adc00df 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -5587,6 +5587,13 @@ static SDValue optimizeIncrementingWhile(SDValue Op, SelectionDAG &DAG,
   SDLoc dl(Op);
   APInt X = Op.getConstantOperandAPInt(1);
   APInt Y = Op.getConstantOperandAPInt(2);
+
+  // When the second operand is the maximum value, comparisons that include
+  // equality can never fail and thus we can return an all active predicate.
+  if (IsEqual)
+    if (IsSigned ? Y.isMaxSignedValue() : Y.isMaxValue())
+      return DAG.getConstant(1, dl, Op.getValueType());
+
   bool Overflow;
   APInt NumActiveElems =
       IsSigned ? Y.ssub_ov(X, Overflow) : Y.usub_ov(X, Overflow);

diff  --git a/llvm/test/CodeGen/AArch64/sve-intrinsics-while.ll b/llvm/test/CodeGen/AArch64/sve-intrinsics-while.ll
index 8380d5ca55cf29..ab4554428be450 100644
--- a/llvm/test/CodeGen/AArch64/sve-intrinsics-while.ll
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsics-while.ll
@@ -128,13 +128,12 @@ define <vscale x 16 x i1> @whilele_b_ii_dont_fold_to_ptrue_overflow() {
   ret <vscale x 16 x i1> %out
 }
 
-define <vscale x 16 x i1> @whilele_b_ii_dont_fold_to_ptrue_increment_overflow() {
-; CHECK-LABEL: whilele_b_ii_dont_fold_to_ptrue_increment_overflow:
+define <vscale x 16 x i1> @whilele_b_ii_known_always_true() {
+; CHECK-LABEL: whilele_b_ii_known_always_true:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #2147483647 // =0x7fffffff
-; CHECK-NEXT:    whilele p0.b, wzr, w8
+; CHECK-NEXT:    ptrue p0.b
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilele.nxv16i1.i32(i32 0, i32 2147483647)
+  %out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilele.nxv16i1.i32(i32 2147483646, i32 2147483647)
   ret <vscale x 16 x i1> %out
 }
 
@@ -388,13 +387,12 @@ define <vscale x 16 x i1> @whilels_b_ii_dont_fold_to_ptrue_overflow() {
   ret <vscale x 16 x i1> %out
 }
 
-define <vscale x 16 x i1> @whilels_b_ii_dont_fold_to_ptrue_increment_overflow() {
-; CHECK-LABEL: whilels_b_ii_dont_fold_to_ptrue_increment_overflow:
+define <vscale x 16 x i1> @whilels_b_ii_known_always_true() {
+; CHECK-LABEL: whilels_b_ii_known_always_true:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
-; CHECK-NEXT:    whilels p0.b, wzr, w8
+; CHECK-NEXT:    ptrue p0.b
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilels.nxv16i1.i32(i32 0, i32 4294967295)
+  %out = call <vscale x 16 x i1> @llvm.aarch64.sve.whilels.nxv16i1.i32(i32 4294967294, i32 4294967295)
   ret <vscale x 16 x i1> %out
 }