[PATCH] D110524: [AArch64ISelLowering] Avoid duplane in some cases when sve enabled

[guopeilin via Phabricator via llvm-commits]

guopeilin updated this revision to Diff 375795.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110524/new/

https://reviews.llvm.org/D110524

Files:
  llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
  llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll


Index: llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll
@@ -0,0 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mattr=+sve -aarch64-sve-vector-bits-min=256 < %s | FileCheck %s
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+define <4 x i32> @test(<16 x i32>* %arg1, <16 x i32>* %arg2) {
+; CHECK-LABEL: test:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT:    sub x9, sp, #48
+; CHECK-NEXT:    mov x29, sp
+; CHECK-NEXT:    and sp, x9, #0xffffffffffffffe0
+; CHECK-NEXT:    .cfi_def_cfa w29, 16
+; CHECK-NEXT:    .cfi_offset w30, -8
+; CHECK-NEXT:    .cfi_offset w29, -16
+; CHECK-NEXT:    mov x8, #8
+; CHECK-NEXT:    ptrue p0.s, vl8
+; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
+; CHECK-NEXT:    mov x9, sp
+; CHECK-NEXT:    ld1w { z1.s }, p0/z, [x0]
+; CHECK-NEXT:    st1w { z0.s }, p0, [x9]
+; CHECK-NEXT:    ldr q2, [sp, #16]
+; CHECK-NEXT:    add z1.s, p0/m, z1.s, z1.s
+; CHECK-NEXT:    movprfx z3, z0
+; CHECK-NEXT:    add z3.s, p0/m, z3.s, z0.s
+; CHECK-NEXT:    st1w { z3.s }, p0, [x0, x8, lsl #2]
+; CHECK-NEXT:    dup v0.4s, v2.s[2]
+; CHECK-NEXT:    st1w { z1.s }, p0, [x0]
+; CHECK-NEXT:    mov sp, x29
+; CHECK-NEXT:    ldp x29, x30, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT:    ret
+entry:
+  %0 = load <16 x i32>, <16 x i32>* %arg1, align 256
+  %1 = load <16 x i32>, <16 x i32>* %arg2, align 256
+  %shvec = shufflevector <16 x i32> %0, <16 x i32> %1, <4 x i32> <i32 14, i32 14, i32 14, i32 14>
+  %2 = add <16 x i32> %0, %0
+  store <16 x i32> %2, <16 x i32>* %arg1, align 256
+  ret <4 x i32> %shvec
+}
+
Index: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -9466,8 +9466,12 @@
   } else if (V.getOpcode() == ISD::EXTRACT_SUBVECTOR) {
     // The lane is incremented by the index of the extract.
     // Example: dup v2f32 (extract v4f32 X, 2), 1 --> dup v4f32 X, 3
-    Lane += V.getConstantOperandVal(1);
-    V = V.getOperand(0);
+    auto ExtractedValType = V.getOperand(0).getValueType();
+    if (ExtractedValType.isFixedLengthVector() &&
+        ExtractedValType.getSizeInBits().getValue() <= 128) {
+      Lane += V.getConstantOperandVal(1);
+      V = V.getOperand(0);
+    }
   } else if (V.getOpcode() == ISD::CONCAT_VECTORS) {
     // The lane is decremented if we are splatting from the 2nd operand.
     // Example: dup v4i32 (concat v2i32 X, v2i32 Y), 3 --> dup v4i32 Y, 1


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D110524.375795.patch
Type: text/x-patch
Size: 2861 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210929/10a1b26d/attachment.bin>