[llvm] 72d3533 - [AArch64][GISel] Clamp bitcast to v2i64 (#191360)

[via llvm-commits]

Author: David Green
Date: 2026-04-13T11:26:29+01:00
New Revision: 72d35333fbd9829f9f89f7f1978ffdbaca396b09

URL: https://github.com/llvm/llvm-project/commit/72d35333fbd9829f9f89f7f1978ffdbaca396b09
DIFF: https://github.com/llvm/llvm-project/commit/72d35333fbd9829f9f89f7f1978ffdbaca396b09.diff

LOG: [AArch64][GISel] Clamp bitcast to v2i64 (#191360)

This helps to not lower, improving the number of nodes that we expand
into and improving the quality of the generated code.

Originally a part of #177158 by Ryan Cowan

Added: 
    

Modified: 
    llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
    llvm/test/CodeGen/AArch64/bitcast.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index cbf0b68629542..5b07480677e47 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -1063,6 +1063,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .clampNumElements(0, v8s8, v16s8)
       .clampNumElements(0, v4s16, v8s16)
       .clampNumElements(0, v2s32, v4s32)
+      .clampMaxNumElements(0, s64, 2)
       .lower();
 
   getActionDefinitionsBuilder(G_VASTART).legalFor({p0});

diff  --git a/llvm/test/CodeGen/AArch64/bitcast.ll b/llvm/test/CodeGen/AArch64/bitcast.ll
index 002e6cd509bec..0f02e1c1f067b 100644
--- a/llvm/test/CodeGen/AArch64/bitcast.ll
+++ b/llvm/test/CodeGen/AArch64/bitcast.ll
@@ -498,10 +498,6 @@ define <4 x i64> @bitcast_v8i32_v4i64(<8 x i32> %a, <8 x i32> %b){
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.4s, v0.4s, v2.4s
 ; CHECK-GI-NEXT:    add v1.4s, v1.4s, v3.4s
-; CHECK-GI-NEXT:    mov x8, v0.d[1]
-; CHECK-GI-NEXT:    mov x9, v1.d[1]
-; CHECK-GI-NEXT:    mov v0.d[1], x8
-; CHECK-GI-NEXT:    mov v1.d[1], x9
 ; CHECK-GI-NEXT:    ret
   %c = add <8 x i32> %a, %b
   %d = bitcast <8 x i32> %c to <4 x i64>
@@ -557,10 +553,6 @@ define <4 x i64> @bitcast_v16i16_v4i64(<16 x i16> %a, <16 x i16> %b){
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.8h, v0.8h, v2.8h
 ; CHECK-GI-NEXT:    add v1.8h, v1.8h, v3.8h
-; CHECK-GI-NEXT:    mov x8, v0.d[1]
-; CHECK-GI-NEXT:    mov x9, v1.d[1]
-; CHECK-GI-NEXT:    mov v0.d[1], x8
-; CHECK-GI-NEXT:    mov v1.d[1], x9
 ; CHECK-GI-NEXT:    ret
   %c = add <16 x i16> %a, %b
   %d = bitcast <16 x i16> %c to <4 x i64>
@@ -599,14 +591,6 @@ define <8 x i64> @bitcast_v16i32_v8i64(<16 x i32> %a, <16 x i32> %b){
 ; CHECK-GI-NEXT:    add v1.4s, v1.4s, v5.4s
 ; CHECK-GI-NEXT:    add v2.4s, v2.4s, v6.4s
 ; CHECK-GI-NEXT:    add v3.4s, v3.4s, v7.4s
-; CHECK-GI-NEXT:    mov x8, v0.d[1]
-; CHECK-GI-NEXT:    mov x9, v1.d[1]
-; CHECK-GI-NEXT:    mov x10, v2.d[1]
-; CHECK-GI-NEXT:    mov x11, v3.d[1]
-; CHECK-GI-NEXT:    mov v0.d[1], x8
-; CHECK-GI-NEXT:    mov v1.d[1], x9
-; CHECK-GI-NEXT:    mov v2.d[1], x10
-; CHECK-GI-NEXT:    mov v3.d[1], x11
 ; CHECK-GI-NEXT:    ret
   %c = add <16 x i32> %a, %b
   %d = bitcast <16 x i32> %c to <8 x i64>