[llvm-branch-commits] [llvm] [AArch64][DAG] Improve codegen for clearing upper vector lanes (PR #237259)

via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Fri Oct 30 08:55:14 PDT 2026


llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-aarch64

Author: Kieran B (kieroxide)

<details>
<summary>Changes</summary>

Add patterns to use FMOV to implicitly clear the upper lanes of a vector while preserving lane 0

---

Patch is 82.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/237259.diff


5 Files Affected:

- (modified) llvm/lib/Target/AArch64/AArch64InstrInfo.td (+41) 
- (modified) llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll (+13-17) 
- (modified) llvm/test/CodeGen/AArch64/build-vector-extract.ll (+1-3) 
- (modified) llvm/test/CodeGen/AArch64/neon-dotreduce.ll (+634-655) 
- (modified) llvm/test/CodeGen/AArch64/shuffle-insert-zero.ll (+91-84) 


``````````diff
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 75739d8cf83ef..14e0d72190398 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -8753,6 +8753,47 @@ defm : Neon_INS_elt_ext_pattern<v16i8, v8i8, v4i16, scalar_to_vector, INSvi8lane
 defm : Neon_INS_elt_ext_pattern<v16i8, v8i8, v2i32, scalar_to_vector_v2f32, INSvi8lane, DUPi8, bsub, VecIndex_x4>;
 defm : Neon_INS_elt_ext_pattern<v8i16, v4i16, v2i32, scalar_to_vector_v2f32, INSvi16lane, DUPi16, hsub, VecIndex_x2>;
 
+multiclass ExtractInsertLane0IntoZeroVector<ValueType VecVT, ValueType ElemTy> {
+  let Predicates = [HasFullFP16] in {
+    def : Pat<(VecVT (vector_insert immAllZerosV,
+                (ElemTy (vector_extract (VecVT V128:$Rn), (i64 0))), (i64 0))),
+              (VecVT (SUBREG_TO_REG (FMOVHr (f16 (EXTRACT_SUBREG (VecVT V128:$Rn), hsub))), hsub))>;
+  }
+}
+
+defm : ExtractInsertLane0IntoZeroVector<v8i16,  i32>;
+defm : ExtractInsertLane0IntoZeroVector<v8f16,  f16>;
+defm : ExtractInsertLane0IntoZeroVector<v8bf16, bf16>;
+
+let Predicates = [HasFullFP16] in {
+  def : Pat<(v4i16 (vector_insert immAllZerosV, (i32 (vector_extract 
+                (v8i16 (insert_subvector (v8i16 undef), (v4i16 V64:$Rn), (i64 0))), 
+                (i64 0))), 
+              (i64 0))),
+            (v4i16 (SUBREG_TO_REG (FMOVHr (f16 (EXTRACT_SUBREG (v4i16 V64:$Rn), hsub))), hsub))>;
+    
+  def : Pat<(v4f16 (vector_insert immAllZerosV, 
+              (f16 (vector_extract 
+                (v8f16 (insert_subvector (v8f16 undef), (v4f16 V64:$Rn), (i64 0))), 
+                (i64 0))), 
+              (i64 0))),
+            (v4f16 (SUBREG_TO_REG (FMOVHr (f16 (EXTRACT_SUBREG (v4f16 V64:$Rn), hsub))), hsub))>;
+        
+  def : Pat<(v4bf16 (vector_insert immAllZerosV, 
+              (bf16 (vector_extract 
+                (v8bf16 (insert_subvector (v8bf16 undef), (v4bf16 V64:$Rn), (i64 0))), 
+                (i64 0))), 
+              (i64 0))),
+            (v4bf16 (SUBREG_TO_REG (FMOVHr (f16 (EXTRACT_SUBREG (v4bf16 V64:$Rn), hsub))), hsub))>;
+}
+
+// extract from lane 0 and insert into lane 0 of a zero vector
+// is equivalent to clearing upper bits of a register, we can use fmov
+def : Pat<(v4i32 (vector_insert immAllZerosV, (i32 (vector_extract (v4i32 V128:$Rn), (i64 0))), (i64 0))),
+          (v4i32 (SUBREG_TO_REG (FMOVSr (EXTRACT_SUBREG (v4i32 V128:$Rn), ssub)), ssub))>;
+def : Pat<(v4f32 (vector_insert immAllZerosV, (f32 (vector_extract (v4f32 V128:$Rn), (i64 0))), (i64 0))),
+          (v4f32 (SUBREG_TO_REG (FMOVSr (EXTRACT_SUBREG (v4f32 V128:$Rn), ssub)), ssub))>;
+
 // bitcast of an extract
 // f32 bitcast(vector_extract(v4i32 src, 0)) -> EXTRACT_SUBREG(src)
 def : Pat<(f32 (bitconvert (i32 (vector_extract v16i8:$src, (i64 0))))),
diff --git a/llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll b/llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll
index 91eda8d552397..ad6fc44e1e7fa 100644
--- a/llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll
+++ b/llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll
@@ -30,8 +30,8 @@ define void @insert_vec_v4i32_uaddlv_from_v8i16(ptr %0) {
 ; CHECK-LABEL: insert_vec_v4i32_uaddlv_from_v8i16:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    movi.2d v0, #0000000000000000
-; CHECK-NEXT:    uaddlv.8h s1, v0
-; CHECK-NEXT:    mov.s v0[0], v1[0]
+; CHECK-NEXT:    uaddlv.8h s0, v0
+; CHECK-NEXT:    fmov s0, s0
 ; CHECK-NEXT:    ucvtf.4s v0, v0
 ; CHECK-NEXT:    str q0, [x0]
 ; CHECK-NEXT:    ret
@@ -48,11 +48,10 @@ define void @insert_vec_v16i32_uaddlv_from_v8i16(ptr %0) {
 ; CHECK-LABEL: insert_vec_v16i32_uaddlv_from_v8i16:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    movi.2d v0, #0000000000000000
-; CHECK-NEXT:    movi.2d v2, #0000000000000000
 ; CHECK-NEXT:    uaddlv.8h s1, v0
 ; CHECK-NEXT:    stp q0, q0, [x0, #32]
-; CHECK-NEXT:    mov.s v2[0], v1[0]
-; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    fmov s1, s1
+; CHECK-NEXT:    ucvtf.4s v1, v1
 ; CHECK-NEXT:    stp q1, q0, [x0]
 ; CHECK-NEXT:    ret
 
@@ -68,14 +67,13 @@ define void @insert_vec_v23i32_uaddlv_from_v8i16(ptr %0) {
 ; CHECK-LABEL: insert_vec_v23i32_uaddlv_from_v8i16:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    movi.2d v0, #0000000000000000
-; CHECK-NEXT:    movi.2d v2, #0000000000000000
 ; CHECK-NEXT:    str wzr, [x0, #88]
 ; CHECK-NEXT:    str xzr, [x0, #80]
 ; CHECK-NEXT:    uaddlv.8h s1, v0
 ; CHECK-NEXT:    stp q0, q0, [x0, #16]
 ; CHECK-NEXT:    stp q0, q0, [x0, #48]
-; CHECK-NEXT:    mov.s v2[0], v1[0]
-; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    fmov s1, s1
+; CHECK-NEXT:    ucvtf.4s v1, v1
 ; CHECK-NEXT:    str q1, [x0]
 ; CHECK-NEXT:    ret
 
@@ -254,10 +252,9 @@ define void @insert_vec_v16i64_uaddlv_from_v4i16(ptr %0) {
 ; CHECK-NEXT:    movi.2d v2, #0000000000000000
 ; CHECK-NEXT:    uaddlv.4h s1, v0
 ; CHECK-NEXT:    stp q0, q0, [x0, #32]
-; CHECK-NEXT:    mov.s v2[0], v1[0]
-; CHECK-NEXT:    fmov x8, d2
-; CHECK-NEXT:    mov.d x9, v2[1]
-; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    fmov s1, s1
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov.d x9, v1[1]
 ; CHECK-NEXT:    ucvtf s1, x8
 ; CHECK-NEXT:    ucvtf s3, x9
 ; CHECK-NEXT:    mov.s v2[0], v1[0]
@@ -346,8 +343,8 @@ define void @insert_vec_v8i32_uaddlv_from_v4i32(ptr %0) {
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    movi.2d v0, #0000000000000000
 ; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
-; CHECK-NEXT:    uaddlv.4s d1, v0
-; CHECK-NEXT:    mov.s v0[0], v1[0]
+; CHECK-NEXT:    uaddlv.4s d0, v0
+; CHECK-NEXT:    fmov s0, s0
 ; CHECK-NEXT:    ucvtf.4s v0, v0
 ; CHECK-NEXT:    str q0, [x0]
 ; CHECK-NEXT:    ret
@@ -365,11 +362,10 @@ define void @insert_vec_v16i32_uaddlv_from_v4i32(ptr %0) {
 ; CHECK-LABEL: insert_vec_v16i32_uaddlv_from_v4i32:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    movi.2d v0, #0000000000000000
-; CHECK-NEXT:    movi.2d v2, #0000000000000000
 ; CHECK-NEXT:    uaddlv.4s d1, v0
 ; CHECK-NEXT:    stp q0, q0, [x0, #32]
-; CHECK-NEXT:    mov.s v2[0], v1[0]
-; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    fmov s1, s1
+; CHECK-NEXT:    ucvtf.4s v1, v1
 ; CHECK-NEXT:    stp q1, q0, [x0]
 ; CHECK-NEXT:    ret
 
diff --git a/llvm/test/CodeGen/AArch64/build-vector-extract.ll b/llvm/test/CodeGen/AArch64/build-vector-extract.ll
index f3e40021b616a..0cfbc939c12a8 100644
--- a/llvm/test/CodeGen/AArch64/build-vector-extract.ll
+++ b/llvm/test/CodeGen/AArch64/build-vector-extract.ll
@@ -15,9 +15,7 @@ define <2 x i64> @extract0_i32_zext_insert0_i64_undef(<4 x i32> %x) {
 define <2 x i64> @extract0_i32_zext_insert0_i64_zero(<4 x i32> %x) {
 ; CHECK-LABEL: extract0_i32_zext_insert0_i64_zero:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-NEXT:    mov v1.s[0], v0.s[0]
-; CHECK-NEXT:    mov v0.16b, v1.16b
+; CHECK-NEXT:    fmov s0, s0
 ; CHECK-NEXT:    ret
   %e = extractelement <4 x i32> %x, i32 0
   %z = zext i32 %e to i64
diff --git a/llvm/test/CodeGen/AArch64/neon-dotreduce.ll b/llvm/test/CodeGen/AArch64/neon-dotreduce.ll
index 2483b476799f6..0fb3cb2643d65 100644
--- a/llvm/test/CodeGen/AArch64/neon-dotreduce.ll
+++ b/llvm/test/CodeGen/AArch64/neon-dotreduce.ll
@@ -321,9 +321,8 @@ define i32 @test_udot_v5i8(ptr nocapture readonly %a, ptr nocapture readonly %b,
 ; CHECK-SD-NEXT:    ldr d0, [x0]
 ; CHECK-SD-NEXT:    ldr d1, [x1]
 ; CHECK-SD-NEXT:    umull v0.8h, v1.8b, v0.8b
-; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-SD-NEXT:    ushll2 v2.4s, v0.8h, #0
-; CHECK-SD-NEXT:    mov v1.s[0], v2.s[0]
+; CHECK-SD-NEXT:    ushll2 v1.4s, v0.8h, #0
+; CHECK-SD-NEXT:    fmov s1, s1
 ; CHECK-SD-NEXT:    uaddw v0.4s, v1.4s, v0.4h
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w8, s0
@@ -406,10 +405,9 @@ define i32 @test_udot_v5i8_nomla(ptr nocapture readonly %a1) {
 ; CHECK-SD-LABEL: test_udot_v5i8_nomla:
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldr d0, [x0]
-; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
 ; CHECK-SD-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-SD-NEXT:    ushll2 v2.4s, v0.8h, #0
-; CHECK-SD-NEXT:    mov v1.s[0], v2.s[0]
+; CHECK-SD-NEXT:    ushll2 v1.4s, v0.8h, #0
+; CHECK-SD-NEXT:    fmov s1, s1
 ; CHECK-SD-NEXT:    uaddw v0.4s, v1.4s, v0.4h
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w0, s0
@@ -447,9 +445,8 @@ define i32 @test_sdot_v5i8(ptr nocapture readonly %a, ptr nocapture readonly %b,
 ; CHECK-SD-NEXT:    ldr d0, [x0]
 ; CHECK-SD-NEXT:    ldr d1, [x1]
 ; CHECK-SD-NEXT:    smull v0.8h, v1.8b, v0.8b
-; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-SD-NEXT:    sshll2 v2.4s, v0.8h, #0
-; CHECK-SD-NEXT:    mov v1.s[0], v2.s[0]
+; CHECK-SD-NEXT:    sshll2 v1.4s, v0.8h, #0
+; CHECK-SD-NEXT:    fmov s1, s1
 ; CHECK-SD-NEXT:    saddw v0.4s, v1.4s, v0.4h
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w8, s0
@@ -533,14 +530,12 @@ define i32 @test_sdot_v5i8_double(<5 x i8> %a, <5 x i8> %b, <5 x i8> %c, <5 x i8
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    smull v2.8h, v2.8b, v3.8b
 ; CHECK-SD-NEXT:    smull v0.8h, v0.8b, v1.8b
-; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-SD-NEXT:    movi v3.2d, #0000000000000000
-; CHECK-SD-NEXT:    sshll2 v4.4s, v0.8h, #0
-; CHECK-SD-NEXT:    sshll2 v5.4s, v2.8h, #0
-; CHECK-SD-NEXT:    mov v3.s[0], v4.s[0]
-; CHECK-SD-NEXT:    mov v1.s[0], v5.s[0]
-; CHECK-SD-NEXT:    saddw v0.4s, v3.4s, v0.4h
+; CHECK-SD-NEXT:    sshll2 v1.4s, v2.8h, #0
+; CHECK-SD-NEXT:    sshll2 v3.4s, v0.8h, #0
+; CHECK-SD-NEXT:    fmov s1, s1
+; CHECK-SD-NEXT:    fmov s3, s3
 ; CHECK-SD-NEXT:    saddw v1.4s, v1.4s, v2.4h
+; CHECK-SD-NEXT:    saddw v0.4s, v3.4s, v0.4h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v1.4s
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w0, s0
@@ -682,14 +677,12 @@ define i32 @test_sdot_v5i8_double_nomla(<5 x i8> %a, <5 x i8> %b, <5 x i8> %c, <
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    sshll v0.8h, v0.8b, #0
 ; CHECK-SD-NEXT:    sshll v1.8h, v2.8b, #0
-; CHECK-SD-NEXT:    movi v2.2d, #0000000000000000
-; CHECK-SD-NEXT:    movi v3.2d, #0000000000000000
-; CHECK-SD-NEXT:    sshll2 v4.4s, v0.8h, #0
-; CHECK-SD-NEXT:    sshll2 v5.4s, v1.8h, #0
-; CHECK-SD-NEXT:    mov v3.s[0], v4.s[0]
-; CHECK-SD-NEXT:    mov v2.s[0], v5.s[0]
-; CHECK-SD-NEXT:    saddw v0.4s, v3.4s, v0.4h
-; CHECK-SD-NEXT:    saddw v1.4s, v2.4s, v1.4h
+; CHECK-SD-NEXT:    sshll2 v2.4s, v0.8h, #0
+; CHECK-SD-NEXT:    sshll2 v3.4s, v1.8h, #0
+; CHECK-SD-NEXT:    fmov s2, s2
+; CHECK-SD-NEXT:    fmov s3, s3
+; CHECK-SD-NEXT:    saddw v0.4s, v2.4s, v0.4h
+; CHECK-SD-NEXT:    saddw v1.4s, v3.4s, v1.4h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v1.4s
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w0, s0
@@ -2310,7 +2303,6 @@ define i32 @test_udot_v25i8(ptr nocapture readonly %a, ptr nocapture readonly %b
 ; CHECK-SD-LABEL: test_udot_v25i8:
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldp q3, q0, [x1]
-; CHECK-SD-NEXT:    movi v5.2d, #0000000000000000
 ; CHECK-SD-NEXT:    ldp q2, q1, [x0]
 ; CHECK-SD-NEXT:    umull2 v4.8h, v0.16b, v1.16b
 ; CHECK-SD-NEXT:    umull v0.8h, v0.8b, v1.8b
@@ -2319,10 +2311,10 @@ define i32 @test_udot_v25i8(ptr nocapture readonly %a, ptr nocapture readonly %b
 ; CHECK-SD-NEXT:    ushll v3.4s, v4.4h, #0
 ; CHECK-SD-NEXT:    uaddl2 v4.4s, v1.8h, v0.8h
 ; CHECK-SD-NEXT:    uaddl v0.4s, v1.4h, v0.4h
-; CHECK-SD-NEXT:    mov v5.s[0], v3.s[0]
+; CHECK-SD-NEXT:    fmov s3, s3
 ; CHECK-SD-NEXT:    uaddw2 v1.4s, v4.4s, v2.8h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v1.4s
-; CHECK-SD-NEXT:    uaddw v2.4s, v5.4s, v2.4h
+; CHECK-SD-NEXT:    uaddw v2.4s, v3.4s, v2.4h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v2.4s
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w8, s0
@@ -2637,20 +2629,19 @@ entry:
 define i32 @test_udot_v25i8_nomla(ptr nocapture readonly %a1) {
 ; CHECK-SD-LABEL: test_udot_v25i8_nomla:
 ; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    ldp q2, q1, [x0]
-; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
-; CHECK-SD-NEXT:    ushll2 v3.8h, v1.16b, #0
-; CHECK-SD-NEXT:    ushll v4.8h, v2.8b, #0
-; CHECK-SD-NEXT:    ushll v1.8h, v1.8b, #0
-; CHECK-SD-NEXT:    ushll2 v2.8h, v2.16b, #0
-; CHECK-SD-NEXT:    ushll v3.4s, v3.4h, #0
-; CHECK-SD-NEXT:    uaddl2 v5.4s, v4.8h, v1.8h
-; CHECK-SD-NEXT:    uaddl v1.4s, v4.4h, v1.4h
-; CHECK-SD-NEXT:    mov v0.s[0], v3.s[0]
-; CHECK-SD-NEXT:    uaddw2 v3.4s, v5.4s, v2.8h
-; CHECK-SD-NEXT:    add v1.4s, v1.4s, v3.4s
-; CHECK-SD-NEXT:    uaddw v0.4s, v0.4s, v2.4h
-; CHECK-SD-NEXT:    add v0.4s, v1.4s, v0.4s
+; CHECK-SD-NEXT:    ldp q1, q0, [x0]
+; CHECK-SD-NEXT:    ushll2 v2.8h, v0.16b, #0
+; CHECK-SD-NEXT:    ushll v3.8h, v1.8b, #0
+; CHECK-SD-NEXT:    ushll v0.8h, v0.8b, #0
+; CHECK-SD-NEXT:    ushll2 v1.8h, v1.16b, #0
+; CHECK-SD-NEXT:    ushll v2.4s, v2.4h, #0
+; CHECK-SD-NEXT:    uaddl2 v4.4s, v3.8h, v0.8h
+; CHECK-SD-NEXT:    uaddl v0.4s, v3.4h, v0.4h
+; CHECK-SD-NEXT:    fmov s2, s2
+; CHECK-SD-NEXT:    uaddw2 v3.4s, v4.4s, v1.8h
+; CHECK-SD-NEXT:    add v0.4s, v0.4s, v3.4s
+; CHECK-SD-NEXT:    uaddw v1.4s, v2.4s, v1.4h
+; CHECK-SD-NEXT:    add v0.4s, v0.4s, v1.4s
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w0, s0
 ; CHECK-SD-NEXT:    ret
@@ -2739,7 +2730,6 @@ define i32 @test_sdot_v25i8(ptr nocapture readonly %a, ptr nocapture readonly %b
 ; CHECK-SD-LABEL: test_sdot_v25i8:
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldp q3, q0, [x1]
-; CHECK-SD-NEXT:    movi v5.2d, #0000000000000000
 ; CHECK-SD-NEXT:    ldp q2, q1, [x0]
 ; CHECK-SD-NEXT:    smull2 v4.8h, v0.16b, v1.16b
 ; CHECK-SD-NEXT:    smull v0.8h, v0.8b, v1.8b
@@ -2748,10 +2738,10 @@ define i32 @test_sdot_v25i8(ptr nocapture readonly %a, ptr nocapture readonly %b
 ; CHECK-SD-NEXT:    sshll v3.4s, v4.4h, #0
 ; CHECK-SD-NEXT:    saddl2 v4.4s, v1.8h, v0.8h
 ; CHECK-SD-NEXT:    saddl v0.4s, v1.4h, v0.4h
-; CHECK-SD-NEXT:    mov v5.s[0], v3.s[0]
+; CHECK-SD-NEXT:    fmov s3, s3
 ; CHECK-SD-NEXT:    saddw2 v1.4s, v4.4s, v2.8h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v1.4s
-; CHECK-SD-NEXT:    saddw v2.4s, v5.4s, v2.4h
+; CHECK-SD-NEXT:    saddw v2.4s, v3.4s, v2.4h
 ; CHECK-SD-NEXT:    add v0.4s, v0.4s, v2.4s
 ; CHECK-SD-NEXT:    addv s0, v0.4s
 ; CHECK-SD-NEXT:    fmov w8, s0
@@ -3073,203 +3063,201 @@ define i32 @test_sdot_v25i8_double(<25 x i8> %a, <25 x i8> %b, <25 x i8> %c, <25
 ; CHECK-SD-NEXT:    add x8, sp, #224
 ; CHECK-SD-NEXT:    ldr b1, [sp, #16]
 ; CHECK-SD-NEXT:    ldr b2, [sp, #280]
-; CHECK-SD-NEXT:    add x9, sp, #240
+; CHECK-SD-NEXT:    add x9, sp, #288
 ; CHECK-SD-NEXT:    ldr b4, [sp, #80]
 ; CHECK-SD-NEXT:    ld1 { v0.b }[1], [x8]
 ; CHECK-SD-NEXT:    add x8, sp, #24
-; CHECK-SD-NEXT:    add x10, sp, #48
+; CHECK-SD-NEXT:    ldr b5, [sp, #152]
 ; CHECK-SD-NEXT:    ld1 { v1.b }[1], [x8]
 ; CHECK-SD-NEXT:    add x8, sp, #232
-; CHECK-SD-NEXT:    add x11, sp, #96
-; CHECK-SD-NEXT:    ldr b5, [sp, #152]
-; CHECK-SD-NEXT:    add x12, sp, #168
-; CHECK-SD-NEXT:    ldr b6, [sp, #616]
+; CHECK-SD-NEXT:    ld1 { v2.b }[1], [x9]
+; CHECK-SD-NEXT:    add x9, sp, #40
+; CHECK-SD-NEXT:    add x10, sp, #296
+; CHECK-SD-NEXT:    add x11, sp, #64
 ; CHECK-SD-NEXT:    ld1 { v0.b }[2], [x8]
 ; CHECK-SD-NEXT:    add x8, sp, #32
-; CHECK-SD-NEXT:    fmov s3, w0
+; CHECK-SD-NEXT:    ldr b6, [sp, #616]
 ; CHECK-SD-NEXT:    ld1 { v1.b }[2], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #288
+; CHECK-SD-NEXT:    add x8, sp, #240
+; CHECK-SD-NEXT:    ld1 { v2.b }[2], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #96
+; CHECK-SD-NEXT:    fmov s3, w0
 ; CHECK-SD-NEXT:    ldr b7, [sp, #416]
-; CHECK-SD-NEXT:    ld1 { v2.b }[1], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #40
-; CHECK-SD-NEXT:    ldr b22, [sp, #744]
-; CHECK-SD-NEXT:    ld1 { v0.b }[3], [x9]
-; CHECK-SD-NEXT:    add x9, sp, #248
-; CHECK-SD-NEXT:    mov v3.b[1], w1
-; CHECK-SD-NEXT:    ld1 { v1.b }[3], [x8]
+; CHECK-SD-NEXT:    ld1 { v0.b }[3], [x8]
 ; CHECK-SD-NEXT:    add x8, sp, #88
-; CHECK-SD-NEXT:    ldr b23, [sp, #544]
+; CHECK-SD-NEXT:    ldr b16, [sp, #344]
+; CHECK-SD-NEXT:    ld1 { v1.b }[3], [x9]
 ; CHECK-SD-NEXT:    ld1 { v4.b }[1], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #256
-; CHECK-SD-NEXT:    ldr b19, [sp, #680]
-; CHECK-SD-NEXT:    ld1 { v0.b }[4], [x9]
-; CHECK-SD-NEXT:    add x9, sp, #296
-; CHECK-SD-NEXT:    ldr b20, [sp, #480]
-; CHECK-SD-NEXT:    ld1 { v1.b }[4], [x10]
-; CHECK-SD-NEXT:    ld1 { v2.b }[2], [x9]
-; CHECK-SD-NEXT:    add x10, sp, #160
-; CHECK-SD-NEXT:    ld1 { v4.b }[2], [x11]
-; CHECK-SD-NEXT:    add x11, sp, #304
-; CHECK-SD-NEXT:    ld1 { v5.b }[1], [x10]
-; CHECK-SD-NEXT:    ld1 { v0.b }[5], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #56
-; CHECK-SD-NEXT:    add x10, sp, #264
-; CHECK-SD-NEXT:    ld1 { v1.b }[5], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #64
-; CHECK-SD-NEXT:    ld1 { v2.b }[3], [x11]
+; CHECK-SD-NEXT:    add x8, sp, #248
+; CHECK-SD-NEXT:    add x9, sp, #160
+; CHECK-SD-NEXT:    mov v3.b[1], w1
+; CHECK-SD-NEXT:    ldr b17, [sp, #144]
+; CHECK-SD-NEXT:    ld1 { v0.b }[4], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #48
+; CHECK-SD-NEXT:    ld1 { v5.b }[1], [x9]
+; CHECK-SD-NEXT:    add x9, sp, #256
+; CHECK-SD-NEXT:    ld1 { v1.b }[4], [x8]
+; CHECK-SD-NEXT:    ld1 { v4.b }[2], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #56
+; CHECK-SD-NEXT:    add x8, sp, #264
+; CHECK-SD-NEXT:    ldr b20, [sp, #744]
+; CHECK-SD-NEXT:    ld1 { v0.b }[5], [x9]
 ; CHECK-SD-NEXT:    add x9, sp, #272
-; CHECK-SD-NEXT:    ld1 { v5.b }[2], [x12]
-; CHECK-SD-NEXT:    add x11, sp, #72
-; CHECK-SD-NEXT:    ld1 { v0.b }[6], [x10]
-; CHECK-SD-NEXT:    add x10, sp, #312
+; CHECK-SD-NEXT:    ldr b21, [sp, #544]
+; CHECK-SD-NEXT:    ld1 { v1.b }[5], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #304
 ; CHECK-SD-NEXT:    mov v3.b[2], w2
-; CHECK-SD-NEXT:    ld1 { v1.b }[6], [x8]
+; CHECK-SD-NEXT:    ld1 { v2.b }[3], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #72
+; CHECK-SD-NEXT:    smull v16.8h, v17.8b, v16.8b
+; CHECK-SD-NEXT:    ld1 { v0.b }[6], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #168
+; CHECK-SD-NEXT:    ldr b17, [sp, #680]
+; CHECK-SD-NEXT:    ld1 { v5.b }[2], [x8]
 ; CHECK-SD-NEXT:    add x8, sp, #104
-; CHECK-SD-NEXT:    ld1 { v2.b }[4], [x10]
+; CHECK-SD-NEXT:    ld1 { v1.b }[6], [x11]
+; CHECK-SD-NEXT:    add x11, sp, #312
 ; CHECK-SD-NEXT:    ld1 { v4.b }[3], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #112
-; CHECK-SD-NEXT:    add x10, sp, #128
+; CHECK-SD-NEXT:    add x8, sp, #320
+; CHECK-SD-NEXT:    ld1 { v2.b }[4], [x11]
 ; CHECK-SD-NEXT:    ld1 { v0.b }[7], [x9]
-; CHECK-SD-NEXT:    add x9, sp, #320
-; CHECK-SD-NEXT:    ldr b21, [sp, #552]
-; CHECK-SD-NEXT:    ld1 { v2.b }[5], [x9]
-; CHECK-SD-NEXT:    add x9, sp, #176
-; CHECK-SD-NEXT:    ld1 { v1.b }[7], [x11]
-; CHECK-SD-NEXT:    ld1 { v4.b }[4], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #624
-; CHECK-SD-NEXT:    ld1 { v5.b }[3], [x9]
-; CHECK-SD-NEXT:    ld1 { v6.b }[1], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #120
-; CHECK-SD-NEXT:    add x9, sp, #328
-; CHECK-SD-NEXT:    ld1 { v2.b }[6], [x9]
-; CHECK-SD-NEXT:    add x9, sp, #184
-; CHECK-SD-NEXT:    add x11, sp, #192
-; CHECK-SD-NEXT:    ld1 { v4.b }[5], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #632
-; CHECK-SD-NEXT:    ld1 { v5.b }[4], [x9]
-; CHECK-SD-NEXT:    ld1 { v6.b }[2], [x8]
-; CHECK-SD-NEXT:    add x9, sp, #640
+; CHECK-SD-NEXT:    add x9, sp, #624
+; CHECK-SD-NEXT:    ld1 { v6.b }[1], [x9]
+; CHECK-SD-NEXT:    ld1 { v1.b }[7], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #112
+; CHECK-SD-NEXT:    add x11, sp, #176
+; CHECK-SD-NEXT:    ld1 { v4.b }[4], [x10]
+; CHECK-SD-NEXT:    add x9, sp, #632
+; CHECK-SD-NEXT:    ld1 { v2.b }[5], [x8]
+; CHECK-SD-NEXT:    ld1 { v5.b }[3], [x11]
+; CHECK-SD-NEXT:    add x8, sp, #328
+; CHECK-SD-NEXT:    ld1 { v6.b }[2], [x9]
+; CHECK-SD-NEXT:    add x10, sp, #120
+; CHECK-SD-NEXT:    add x11, sp, #184
+; CHECK-SD-NEXT:    ld1 { v4.b }[5], [x10]
+; CHECK-SD-NEXT:    add x10, sp, #640
+; CHECK-SD-NEXT:    add x9, sp, #128
+; CHECK-SD-NEXT:    ld1 { v2.b }[6], [x8]
+; CHECK-SD-NEXT:    ld1 { v5.b }[4], [x11]
 ; CHECK-SD-NEXT:    add x8, sp, #336
+; CHECK-SD-NEXT:    ld1 { v6.b }[3], [x10]
+; CHECK-SD-NEXT:    add x11, sp, #192
+; CHECK-SD-NEXT:    ldr b18, [sp, #480]
+; CHECK-SD-NEXT:    ld1 { v4.b }[6], [x9]
+; CHECK-SD-NEXT:    add x9, sp, #200
+; CHECK-SD-NEXT:    ldr b19, [sp, #552]
 ; CHECK-SD-NEXT:    ld1 { v2.b }[7], [x8]
-; CHECK-SD-NEXT:    add x8, sp, #656
-; CHECK-SD-NEXT:...
[truncated]

``````````

</details>


https://github.com/llvm/llvm-project/pull/237259


More information about the llvm-branch-commits mailing list