[llvm] b9ecf45 - [AArch64][GlobalISel] Update and cleanup s/umul.fix tests. NFC (#211232)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jul 22 04:05:59 PDT 2026
Author: David Green
Date: 2026-07-22T12:05:54+01:00
New Revision: b9ecf4590babac78a761efd1704ceefd4a1a26e1
URL: https://github.com/llvm/llvm-project/commit/b9ecf4590babac78a761efd1704ceefd4a1a26e1
DIFF: https://github.com/llvm/llvm-project/commit/b9ecf4590babac78a761efd1704ceefd4a1a26e1.diff
LOG: [AArch64][GlobalISel] Update and cleanup s/umul.fix tests. NFC (#211232)
Added:
Modified:
llvm/test/CodeGen/AArch64/smul_fix.ll
llvm/test/CodeGen/AArch64/smul_fix_sat.ll
llvm/test/CodeGen/AArch64/umul_fix.ll
llvm/test/CodeGen/AArch64/umul_fix_sat.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/AArch64/smul_fix.ll b/llvm/test/CodeGen/AArch64/smul_fix.ll
index 325d9d3d02cc7..b2f8472234d40 100644
--- a/llvm/test/CodeGen/AArch64/smul_fix.ll
+++ b/llvm/test/CodeGen/AArch64/smul_fix.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=0 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
-; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-define i32 @func(i32 %x, i32 %y) nounwind {
+define i32 @func(i32 %x, i32 %y) {
; CHECK-SD-LABEL: func:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: smull x8, w0, w1
@@ -42,7 +42,7 @@ define i64 @func2(i64 %x, i64 %y) {
ret i64 %tmp
}
-define i4 @func3(i4 %x, i4 %y) nounwind {
+define i4 @func3(i4 %x, i4 %y) {
; CHECK-SD-LABEL: func3:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: sbfx w8, w1, #0, #4
@@ -64,7 +64,7 @@ define i4 @func3(i4 %x, i4 %y) nounwind {
}
;; These result in regular integer multiplication
-define i32 @func4(i32 %x, i32 %y) nounwind {
+define i32 @func4(i32 %x, i32 %y) {
; CHECK-LABEL: func4:
; CHECK: // %bb.0:
; CHECK-NEXT: mul w0, w0, w1
@@ -82,7 +82,7 @@ define i64 @func5(i64 %x, i64 %y) {
ret i64 %tmp
}
-define i4 @func6(i4 %x, i4 %y) nounwind {
+define i4 @func6(i4 %x, i4 %y) {
; CHECK-SD-LABEL: func6:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: sbfx w8, w1, #0, #4
@@ -98,7 +98,7 @@ define i4 @func6(i4 %x, i4 %y) nounwind {
ret i4 %tmp
}
-define i64 @func7(i64 %x, i64 %y) nounwind {
+define i64 @func7(i64 %x, i64 %y) {
; CHECK-SD-LABEL: func7:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mul x8, x0, x1
@@ -120,7 +120,7 @@ define i64 @func7(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i64 @func8(i64 %x, i64 %y) nounwind {
+define i64 @func8(i64 %x, i64 %y) {
; CHECK-SD-LABEL: func8:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mul x8, x0, x1
@@ -142,8 +142,90 @@ define i64 @func8(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define <2 x i32> @smulfix_2xi32_0(<2 x i32> %x, <2 x i32> %y) nounwind {
-; CHECK-LABEL: smulfix_2xi32_0:
+define <8 x i8> @vec_v8i8(<8 x i8> %x, <8 x i8> %y) {
+; CHECK-SD-LABEL: vec_v8i8:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: smull v0.8h, v0.8b, v1.8b
+; CHECK-SD-NEXT: shrn v1.8b, v0.8h, #8
+; CHECK-SD-NEXT: xtn v2.8b, v0.8h
+; CHECK-SD-NEXT: shl v0.8b, v1.8b, #6
+; CHECK-SD-NEXT: usra v0.8b, v2.8b, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i8:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: smull v0.8h, v0.8b, v1.8b
+; CHECK-GI-NEXT: shrn v0.8b, v0.8h, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i8> @llvm.smul.fix.v8i8(<8 x i8> %x, <8 x i8> %y, i32 2)
+ ret <8 x i8> %tmp
+}
+
+define <16 x i8> @vec_v16i8(<16 x i8> %x, <16 x i8> %y) {
+; CHECK-SD-LABEL: vec_v16i8:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: smull2 v2.8h, v0.16b, v1.16b
+; CHECK-SD-NEXT: smull v3.8h, v0.8b, v1.8b
+; CHECK-SD-NEXT: mul v1.16b, v0.16b, v1.16b
+; CHECK-SD-NEXT: uzp2 v2.16b, v3.16b, v2.16b
+; CHECK-SD-NEXT: shl v0.16b, v2.16b, #6
+; CHECK-SD-NEXT: usra v0.16b, v1.16b, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v16i8:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: smull v2.8h, v0.8b, v1.8b
+; CHECK-GI-NEXT: smull2 v1.8h, v0.16b, v1.16b
+; CHECK-GI-NEXT: shrn v0.8b, v2.8h, #2
+; CHECK-GI-NEXT: shrn2 v0.16b, v1.8h, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> %x, <16 x i8> %y, i32 2)
+ ret <16 x i8> %tmp
+}
+
+define <4 x i16> @vec_v4i16(<4 x i16> %x, <4 x i16> %y) {
+; CHECK-SD-LABEL: vec_v4i16:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: smull v0.4s, v0.4h, v1.4h
+; CHECK-SD-NEXT: shrn v1.4h, v0.4s, #16
+; CHECK-SD-NEXT: xtn v2.4h, v0.4s
+; CHECK-SD-NEXT: shl v0.4h, v1.4h, #6
+; CHECK-SD-NEXT: usra v0.4h, v2.4h, #10
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v4i16:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: smull v0.4s, v0.4h, v1.4h
+; CHECK-GI-NEXT: shrn v0.4h, v0.4s, #10
+; CHECK-GI-NEXT: ret
+ %tmp = call <4 x i16> @llvm.smul.fix.v4i16(<4 x i16> %x, <4 x i16> %y, i32 10)
+ ret <4 x i16> %tmp
+}
+
+define <8 x i16> @vec_v8i16(<8 x i16> %x, <8 x i16> %y) {
+; CHECK-SD-LABEL: vec_v8i16:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: smull2 v2.4s, v0.8h, v1.8h
+; CHECK-SD-NEXT: smull v3.4s, v0.4h, v1.4h
+; CHECK-SD-NEXT: mul v1.8h, v0.8h, v1.8h
+; CHECK-SD-NEXT: uzp2 v2.8h, v3.8h, v2.8h
+; CHECK-SD-NEXT: shl v0.8h, v2.8h, #14
+; CHECK-SD-NEXT: usra v0.8h, v1.8h, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i16:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: smull v2.4s, v0.4h, v1.4h
+; CHECK-GI-NEXT: smull2 v1.4s, v0.8h, v1.8h
+; CHECK-GI-NEXT: shrn v0.4h, v2.4s, #2
+; CHECK-GI-NEXT: shrn2 v0.8h, v1.4s, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> %x, <8 x i16> %y, i32 2)
+ ret <8 x i16> %tmp
+}
+
+define <2 x i32> @vec_v2i32(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: vec_v2i32:
; CHECK: // %bb.0:
; CHECK-NEXT: mul v0.2s, v0.2s, v1.2s
; CHECK-NEXT: ret
@@ -151,17 +233,8 @@ define <2 x i32> @smulfix_2xi32_0(<2 x i32> %x, <2 x i32> %y) nounwind {
ret <2 x i32> %tmp
}
-define <4 x i32> @smulfix_4xi32_0(<4 x i32> %x, <4 x i32> %y) nounwind {
-; CHECK-LABEL: smulfix_4xi32_0:
-; CHECK: // %bb.0:
-; CHECK-NEXT: mul v0.4s, v0.4s, v1.4s
-; CHECK-NEXT: ret
- %tmp = call <4 x i32> @llvm.smul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 0)
- ret <4 x i32> %tmp
-}
-
-define <4 x i32> @smulfix_4xi32(<4 x i32> %1, <4 x i32> %2) {
-; CHECK-SD-LABEL: smulfix_4xi32:
+define <4 x i32> @vec_v4i32(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-SD-LABEL: vec_v4i32:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: smull2 v2.2d, v0.4s, v1.4s
; CHECK-SD-NEXT: smull v3.2d, v0.2s, v1.2s
@@ -171,19 +244,81 @@ define <4 x i32> @smulfix_4xi32(<4 x i32> %1, <4 x i32> %2) {
; CHECK-SD-NEXT: usra v0.4s, v1.4s, #15
; CHECK-SD-NEXT: ret
;
-; CHECK-GI-LABEL: smulfix_4xi32:
+; CHECK-GI-LABEL: vec_v4i32:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: smull v2.2d, v0.2s, v1.2s
; CHECK-GI-NEXT: smull2 v1.2d, v0.4s, v1.4s
; CHECK-GI-NEXT: shrn v0.2s, v2.2d, #15
; CHECK-GI-NEXT: shrn2 v0.4s, v1.2d, #15
; CHECK-GI-NEXT: ret
- %m = call <4 x i32> @llvm.smul.fix.v4i32(<4 x i32> %1, <4 x i32> %2, i32 15)
- ret <4 x i32> %m
+ %tmp = call <4 x i32> @llvm.smul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 15)
+ ret <4 x i32> %tmp
+}
+
+define <8 x i32> @vec_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; CHECK-SD-LABEL: vec_v8i32:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: mul v1.4s, v1.4s, v3.4s
+; CHECK-SD-NEXT: mul v0.4s, v0.4s, v2.4s
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i32:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: mul v0.4s, v0.4s, v2.4s
+; CHECK-GI-NEXT: mul v1.4s, v1.4s, v3.4s
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i32> @llvm.smul.fix.v8i32(<8 x i32> %x, <8 x i32> %y, i32 0)
+ ret <8 x i32> %tmp
+}
+
+define <2 x i64> @vec_v2i64(<2 x i64> %x, <2 x i64> %y) {
+; CHECK-SD-LABEL: vec_v2i64:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: fmov x8, d1
+; CHECK-SD-NEXT: fmov x9, d0
+; CHECK-SD-NEXT: mov x10, v1.d[1]
+; CHECK-SD-NEXT: mov x11, v0.d[1]
+; CHECK-SD-NEXT: mul x12, x9, x8
+; CHECK-SD-NEXT: smulh x8, x9, x8
+; CHECK-SD-NEXT: mul x9, x11, x10
+; CHECK-SD-NEXT: smulh x10, x11, x10
+; CHECK-SD-NEXT: extr x8, x8, x12, #15
+; CHECK-SD-NEXT: fmov d0, x8
+; CHECK-SD-NEXT: extr x9, x10, x9, #15
+; CHECK-SD-NEXT: mov v0.d[1], x9
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v2i64:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: fmov x8, d0
+; CHECK-GI-NEXT: fmov x9, d1
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: mov d1, v1.d[1]
+; CHECK-GI-NEXT: umulh x10, x8, x9
+; CHECK-GI-NEXT: asr x11, x9, #63
+; CHECK-GI-NEXT: fmov x13, d1
+; CHECK-GI-NEXT: mul x12, x8, x9
+; CHECK-GI-NEXT: madd x10, x8, x11, x10
+; CHECK-GI-NEXT: fmov x11, d0
+; CHECK-GI-NEXT: asr x8, x8, #63
+; CHECK-GI-NEXT: asr x15, x13, #63
+; CHECK-GI-NEXT: umulh x14, x11, x13
+; CHECK-GI-NEXT: madd x8, x8, x9, x10
+; CHECK-GI-NEXT: asr x10, x11, #63
+; CHECK-GI-NEXT: madd x9, x11, x15, x14
+; CHECK-GI-NEXT: mul x11, x11, x13
+; CHECK-GI-NEXT: extr x8, x8, x12, #15
+; CHECK-GI-NEXT: madd x9, x10, x13, x9
+; CHECK-GI-NEXT: fmov d0, x8
+; CHECK-GI-NEXT: extr x9, x9, x11, #15
+; CHECK-GI-NEXT: mov v0.d[1], x9
+; CHECK-GI-NEXT: ret
+ %tmp = call <2 x i64> @llvm.smul.fix.v2i64(<2 x i64> %x, <2 x i64> %y, i32 15)
+ ret <2 x i64> %tmp
}
-define <4 x i64> @smulfix_4xi64(<4 x i64> %x, <4 x i64> %y) nounwind {
-; CHECK-SD-LABEL: smulfix_4xi64:
+define <4 x i64> @vec_v4i64(<4 x i64> %x, <4 x i64> %y) {
+; CHECK-SD-LABEL: vec_v4i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov x8, v2.d[1]
; CHECK-SD-NEXT: mov x9, v0.d[1]
@@ -211,7 +346,7 @@ define <4 x i64> @smulfix_4xi64(<4 x i64> %x, <4 x i64> %y) nounwind {
; CHECK-SD-NEXT: mov v1.d[1], x11
; CHECK-SD-NEXT: ret
;
-; CHECK-GI-LABEL: smulfix_4xi64:
+; CHECK-GI-LABEL: vec_v4i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: fmov x8, d0
; CHECK-GI-NEXT: fmov x9, d2
@@ -261,22 +396,3 @@ define <4 x i64> @smulfix_4xi64(<4 x i64> %x, <4 x i64> %y) nounwind {
%tmp = call <4 x i64> @llvm.smul.fix.v4i64(<4 x i64> %x, <4 x i64> %y, i32 32)
ret <4 x i64> %tmp
}
-
-define <4 x i16> @widemul(<4 x i16> %x, <4 x i16> %y) nounwind {
-; CHECK-SD-LABEL: widemul:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: smull v0.4s, v0.4h, v1.4h
-; CHECK-SD-NEXT: shrn v1.4h, v0.4s, #16
-; CHECK-SD-NEXT: xtn v2.4h, v0.4s
-; CHECK-SD-NEXT: shl v0.4h, v1.4h, #14
-; CHECK-SD-NEXT: usra v0.4h, v2.4h, #2
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: widemul:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: smull v0.4s, v0.4h, v1.4h
-; CHECK-GI-NEXT: shrn v0.4h, v0.4s, #2
-; CHECK-GI-NEXT: ret
- %tmp = call <4 x i16> @llvm.smul.fix.v4i16(<4 x i16> %x, <4 x i16> %y, i32 2)
- ret <4 x i16> %tmp
-}
diff --git a/llvm/test/CodeGen/AArch64/smul_fix_sat.ll b/llvm/test/CodeGen/AArch64/smul_fix_sat.ll
index 7cb680b8449cf..4cef8160da145 100644
--- a/llvm/test/CodeGen/AArch64/smul_fix_sat.ll
+++ b/llvm/test/CodeGen/AArch64/smul_fix_sat.ll
@@ -1,7 +1,26 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=aarch64-linux-gnu | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=0 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=1 -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-define i32 @func(i32 %x, i32 %y) nounwind {
+; CHECK-GI: warning: Instruction selection used fallback path for func
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func2
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func3
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func4
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func5
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func6
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func7
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v16i8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i16
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i16
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v2i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v2i64
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i64
+
+define i32 @func(i32 %x, i32 %y) {
; CHECK-LABEL: func:
; CHECK: // %bb.0:
; CHECK-NEXT: smull x9, w0, w1
@@ -18,7 +37,7 @@ define i32 @func(i32 %x, i32 %y) nounwind {
ret i32 %tmp
}
-define i64 @func2(i64 %x, i64 %y) nounwind {
+define i64 @func2(i64 %x, i64 %y) {
; CHECK-LABEL: func2:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x9, x0, x1
@@ -35,7 +54,7 @@ define i64 @func2(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i4 @func3(i4 %x, i4 %y) nounwind {
+define i4 @func3(i4 %x, i4 %y) {
; CHECK-LABEL: func3:
; CHECK: // %bb.0:
; CHECK-NEXT: sbfx w9, w1, #0, #4
@@ -56,7 +75,7 @@ define i4 @func3(i4 %x, i4 %y) nounwind {
}
;; These result in regular integer multiplication with a saturation check.
-define i32 @func4(i32 %x, i32 %y) nounwind {
+define i32 @func4(i32 %x, i32 %y) {
; CHECK-LABEL: func4:
; CHECK: // %bb.0:
; CHECK-NEXT: smull x9, w0, w1
@@ -87,7 +106,7 @@ define i64 @func5(i64 %x, i64 %y) {
ret i64 %tmp
}
-define i4 @func6(i4 %x, i4 %y) nounwind {
+define i4 @func6(i4 %x, i4 %y) {
; CHECK-LABEL: func6:
; CHECK: // %bb.0:
; CHECK-NEXT: sbfx w9, w1, #0, #4
@@ -105,7 +124,7 @@ define i4 @func6(i4 %x, i4 %y) nounwind {
ret i4 %tmp
}
-define i64 @func7(i64 %x, i64 %y) nounwind {
+define i64 @func7(i64 %x, i64 %y) {
; CHECK-LABEL: func7:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x9, x0, x1
@@ -124,7 +143,7 @@ define i64 @func7(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i64 @func8(i64 %x, i64 %y) nounwind {
+define i64 @func8(i64 %x, i64 %y) {
; CHECK-LABEL: func8:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x9, x0, x1
@@ -143,8 +162,488 @@ define i64 @func8(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
-; CHECK-LABEL: vec:
+define <8 x i8> @vec_v8i8(<8 x i8> %x, <8 x i8> %y) {
+; CHECK-LABEL: vec_v8i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: umov w8, v0.b[0]
+; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NEXT: smov x9, v1.b[0]
+; CHECK-NEXT: umov w10, v0.b[1]
+; CHECK-NEXT: smov x11, v1.b[1]
+; CHECK-NEXT: umov w12, v0.b[2]
+; CHECK-NEXT: smov x14, v1.b[2]
+; CHECK-NEXT: umov w15, v0.b[3]
+; CHECK-NEXT: smov x16, v1.b[3]
+; CHECK-NEXT: umov w17, v0.b[4]
+; CHECK-NEXT: lsl w8, w8, #24
+; CHECK-NEXT: lsl w10, w10, #24
+; CHECK-NEXT: smull x13, w9, w8
+; CHECK-NEXT: mov w9, #2147483647 // =0x7fffffff
+; CHECK-NEXT: mov w8, #-2147483648 // =0x80000000
+; CHECK-NEXT: smull x10, w11, w10
+; CHECK-NEXT: lsl w11, w12, #24
+; CHECK-NEXT: lsl w15, w15, #24
+; CHECK-NEXT: lsl w17, w17, #24
+; CHECK-NEXT: lsr x12, x13, #32
+; CHECK-NEXT: smull x11, w14, w11
+; CHECK-NEXT: lsr x14, x10, #32
+; CHECK-NEXT: smull x15, w16, w15
+; CHECK-NEXT: extr w13, w12, w13, #2
+; CHECK-NEXT: cmp w12, #1
+; CHECK-NEXT: extr w10, w14, w10, #2
+; CHECK-NEXT: lsr x16, x11, #32
+; CHECK-NEXT: csel w13, w9, w13, gt
+; CHECK-NEXT: cmn w12, #2
+; CHECK-NEXT: smov x12, v1.b[4]
+; CHECK-NEXT: csel w13, w8, w13, lt
+; CHECK-NEXT: cmp w14, #1
+; CHECK-NEXT: extr w11, w16, w11, #2
+; CHECK-NEXT: asr w13, w13, #24
+; CHECK-NEXT: csel w10, w9, w10, gt
+; CHECK-NEXT: cmn w14, #2
+; CHECK-NEXT: umov w14, v0.b[5]
+; CHECK-NEXT: csel w10, w8, w10, lt
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: fmov s2, w13
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: smov x13, v1.b[5]
+; CHECK-NEXT: smull x12, w12, w17
+; CHECK-NEXT: csel w11, w9, w11, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: lsr x16, x15, #32
+; CHECK-NEXT: csel w11, w8, w11, lt
+; CHECK-NEXT: mov v2.b[1], w10
+; CHECK-NEXT: lsl w10, w14, #24
+; CHECK-NEXT: asr w11, w11, #24
+; CHECK-NEXT: extr w15, w16, w15, #2
+; CHECK-NEXT: umov w14, v0.b[6]
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: smull x10, w13, w10
+; CHECK-NEXT: lsr x13, x12, #32
+; CHECK-NEXT: csel w15, w9, w15, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: mov v2.b[2], w11
+; CHECK-NEXT: smov x11, v1.b[6]
+; CHECK-NEXT: extr w12, w13, w12, #2
+; CHECK-NEXT: csel w15, w8, w15, lt
+; CHECK-NEXT: cmp w13, #1
+; CHECK-NEXT: lsl w14, w14, #24
+; CHECK-NEXT: asr w15, w15, #24
+; CHECK-NEXT: csel w12, w9, w12, gt
+; CHECK-NEXT: cmn w13, #2
+; CHECK-NEXT: umov w13, v0.b[7]
+; CHECK-NEXT: lsr x16, x10, #32
+; CHECK-NEXT: csel w12, w8, w12, lt
+; CHECK-NEXT: mov v2.b[3], w15
+; CHECK-NEXT: smull x11, w11, w14
+; CHECK-NEXT: smov x14, v1.b[7]
+; CHECK-NEXT: extr w10, w16, w10, #2
+; CHECK-NEXT: asr w12, w12, #24
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: lsl w13, w13, #24
+; CHECK-NEXT: csel w10, w9, w10, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: mov v2.b[4], w12
+; CHECK-NEXT: lsr x12, x11, #32
+; CHECK-NEXT: csel w10, w8, w10, lt
+; CHECK-NEXT: smull x13, w14, w13
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: extr w11, w12, w11, #2
+; CHECK-NEXT: cmp w12, #1
+; CHECK-NEXT: mov v2.b[5], w10
+; CHECK-NEXT: csel w10, w9, w11, gt
+; CHECK-NEXT: cmn w12, #2
+; CHECK-NEXT: lsr x11, x13, #32
+; CHECK-NEXT: csel w10, w8, w10, lt
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: extr w12, w11, w13, #2
+; CHECK-NEXT: cmp w11, #1
+; CHECK-NEXT: mov v2.b[6], w10
+; CHECK-NEXT: csel w9, w9, w12, gt
+; CHECK-NEXT: cmn w11, #2
+; CHECK-NEXT: csel w8, w8, w9, lt
+; CHECK-NEXT: asr w8, w8, #24
+; CHECK-NEXT: mov v2.b[7], w8
+; CHECK-NEXT: fmov d0, d2
+; CHECK-NEXT: ret
+ %tmp = call <8 x i8> @llvm.smul.fix.sat.v8i8(<8 x i8> %x, <8 x i8> %y, i32 2)
+ ret <8 x i8> %tmp
+}
+
+define <16 x i8> @vec_v16i8(<16 x i8> %x, <16 x i8> %y) {
+; CHECK-LABEL: vec_v16i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: umov w9, v0.b[1]
+; CHECK-NEXT: smov x10, v1.b[1]
+; CHECK-NEXT: mov w8, #2147483647 // =0x7fffffff
+; CHECK-NEXT: umov w11, v0.b[0]
+; CHECK-NEXT: smov x12, v1.b[0]
+; CHECK-NEXT: umov w13, v0.b[2]
+; CHECK-NEXT: umov w15, v0.b[3]
+; CHECK-NEXT: umov w16, v0.b[4]
+; CHECK-NEXT: smov x17, v1.b[3]
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: lsl w13, w13, #24
+; CHECK-NEXT: smull x10, w10, w9
+; CHECK-NEXT: lsl w9, w11, #24
+; CHECK-NEXT: smov x11, v1.b[2]
+; CHECK-NEXT: lsl w15, w15, #24
+; CHECK-NEXT: lsl w16, w16, #24
+; CHECK-NEXT: smull x12, w12, w9
+; CHECK-NEXT: mov w9, #-2147483648 // =0x80000000
+; CHECK-NEXT: lsr x14, x10, #32
+; CHECK-NEXT: smull x15, w17, w15
+; CHECK-NEXT: umov w17, v0.b[5]
+; CHECK-NEXT: smull x11, w11, w13
+; CHECK-NEXT: extr w10, w14, w10, #2
+; CHECK-NEXT: lsr x13, x12, #32
+; CHECK-NEXT: cmp w14, #1
+; CHECK-NEXT: csel w10, w8, w10, gt
+; CHECK-NEXT: cmn w14, #2
+; CHECK-NEXT: extr w12, w13, w12, #2
+; CHECK-NEXT: csel w10, w9, w10, lt
+; CHECK-NEXT: cmp w13, #1
+; CHECK-NEXT: smov x14, v1.b[4]
+; CHECK-NEXT: csel w12, w8, w12, gt
+; CHECK-NEXT: cmn w13, #2
+; CHECK-NEXT: lsr x13, x11, #32
+; CHECK-NEXT: csel w12, w9, w12, lt
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: asr w12, w12, #24
+; CHECK-NEXT: extr w11, w13, w11, #2
+; CHECK-NEXT: cmp w13, #1
+; CHECK-NEXT: smull x14, w14, w16
+; CHECK-NEXT: smov x16, v1.b[5]
+; CHECK-NEXT: fmov s2, w12
+; CHECK-NEXT: lsr x12, x15, #32
+; CHECK-NEXT: csel w11, w8, w11, gt
+; CHECK-NEXT: cmn w13, #2
+; CHECK-NEXT: umov w13, v0.b[6]
+; CHECK-NEXT: csel w11, w9, w11, lt
+; CHECK-NEXT: extr w15, w12, w15, #2
+; CHECK-NEXT: cmp w12, #1
+; CHECK-NEXT: mov v2.b[1], w10
+; CHECK-NEXT: lsl w10, w17, #24
+; CHECK-NEXT: asr w11, w11, #24
+; CHECK-NEXT: smov x17, v1.b[6]
+; CHECK-NEXT: csel w15, w8, w15, gt
+; CHECK-NEXT: cmn w12, #2
+; CHECK-NEXT: smull x10, w16, w10
+; CHECK-NEXT: lsr x16, x14, #32
+; CHECK-NEXT: umov w12, v0.b[7]
+; CHECK-NEXT: mov v2.b[2], w11
+; CHECK-NEXT: lsl w11, w13, #24
+; CHECK-NEXT: extr w13, w16, w14, #2
+; CHECK-NEXT: csel w14, w9, w15, lt
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: smov x15, v1.b[7]
+; CHECK-NEXT: asr w14, w14, #24
+; CHECK-NEXT: smull x11, w17, w11
+; CHECK-NEXT: lsr x17, x10, #32
+; CHECK-NEXT: csel w13, w8, w13, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: umov w16, v0.b[8]
+; CHECK-NEXT: mov v2.b[3], w14
+; CHECK-NEXT: csel w13, w9, w13, lt
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: extr w10, w17, w10, #2
+; CHECK-NEXT: asr w13, w13, #24
+; CHECK-NEXT: cmp w17, #1
+; CHECK-NEXT: smov x14, v1.b[8]
+; CHECK-NEXT: smull x12, w15, w12
+; CHECK-NEXT: lsr x15, x11, #32
+; CHECK-NEXT: csel w10, w8, w10, gt
+; CHECK-NEXT: cmn w17, #2
+; CHECK-NEXT: umov w17, v0.b[9]
+; CHECK-NEXT: mov v2.b[4], w13
+; CHECK-NEXT: csel w10, w9, w10, lt
+; CHECK-NEXT: lsl w13, w16, #24
+; CHECK-NEXT: extr w11, w15, w11, #2
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: cmp w15, #1
+; CHECK-NEXT: smov x16, v1.b[9]
+; CHECK-NEXT: smull x13, w14, w13
+; CHECK-NEXT: lsr x14, x12, #32
+; CHECK-NEXT: csel w11, w8, w11, gt
+; CHECK-NEXT: cmn w15, #2
+; CHECK-NEXT: umov w15, v0.b[10]
+; CHECK-NEXT: mov v2.b[5], w10
+; CHECK-NEXT: csel w11, w9, w11, lt
+; CHECK-NEXT: lsl w10, w17, #24
+; CHECK-NEXT: extr w12, w14, w12, #2
+; CHECK-NEXT: asr w11, w11, #24
+; CHECK-NEXT: cmp w14, #1
+; CHECK-NEXT: smov x17, v1.b[10]
+; CHECK-NEXT: smull x10, w16, w10
+; CHECK-NEXT: lsr x16, x13, #32
+; CHECK-NEXT: csel w12, w8, w12, gt
+; CHECK-NEXT: cmn w14, #2
+; CHECK-NEXT: umov w14, v0.b[11]
+; CHECK-NEXT: mov v2.b[6], w11
+; CHECK-NEXT: csel w12, w9, w12, lt
+; CHECK-NEXT: lsl w11, w15, #24
+; CHECK-NEXT: extr w13, w16, w13, #2
+; CHECK-NEXT: asr w12, w12, #24
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: smov x15, v1.b[11]
+; CHECK-NEXT: smull x11, w17, w11
+; CHECK-NEXT: lsr x17, x10, #32
+; CHECK-NEXT: csel w13, w8, w13, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: umov w16, v0.b[12]
+; CHECK-NEXT: mov v2.b[7], w12
+; CHECK-NEXT: csel w13, w9, w13, lt
+; CHECK-NEXT: lsl w12, w14, #24
+; CHECK-NEXT: extr w10, w17, w10, #2
+; CHECK-NEXT: asr w13, w13, #24
+; CHECK-NEXT: cmp w17, #1
+; CHECK-NEXT: smull x12, w15, w12
+; CHECK-NEXT: lsr x15, x11, #32
+; CHECK-NEXT: smov x14, v1.b[12]
+; CHECK-NEXT: csel w10, w8, w10, gt
+; CHECK-NEXT: cmn w17, #2
+; CHECK-NEXT: lsl w16, w16, #24
+; CHECK-NEXT: mov v2.b[8], w13
+; CHECK-NEXT: csel w10, w9, w10, lt
+; CHECK-NEXT: extr w11, w15, w11, #2
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: umov w13, v0.b[13]
+; CHECK-NEXT: cmp w15, #1
+; CHECK-NEXT: csel w11, w8, w11, gt
+; CHECK-NEXT: cmn w15, #2
+; CHECK-NEXT: lsr x15, x12, #32
+; CHECK-NEXT: smull x14, w14, w16
+; CHECK-NEXT: smov x16, v1.b[13]
+; CHECK-NEXT: csel w11, w9, w11, lt
+; CHECK-NEXT: mov v2.b[9], w10
+; CHECK-NEXT: extr w12, w15, w12, #2
+; CHECK-NEXT: asr w11, w11, #24
+; CHECK-NEXT: lsl w10, w13, #24
+; CHECK-NEXT: cmp w15, #1
+; CHECK-NEXT: umov w13, v0.b[14]
+; CHECK-NEXT: csel w12, w8, w12, gt
+; CHECK-NEXT: cmn w15, #2
+; CHECK-NEXT: smull x10, w16, w10
+; CHECK-NEXT: lsr x16, x14, #32
+; CHECK-NEXT: csel w12, w9, w12, lt
+; CHECK-NEXT: mov v2.b[10], w11
+; CHECK-NEXT: smov x11, v1.b[14]
+; CHECK-NEXT: asr w12, w12, #24
+; CHECK-NEXT: extr w14, w16, w14, #2
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: lsl w13, w13, #24
+; CHECK-NEXT: lsr x15, x10, #32
+; CHECK-NEXT: csel w14, w8, w14, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: mov v2.b[11], w12
+; CHECK-NEXT: umov w12, v0.b[15]
+; CHECK-NEXT: smull x11, w11, w13
+; CHECK-NEXT: csel w13, w9, w14, lt
+; CHECK-NEXT: smov x14, v1.b[15]
+; CHECK-NEXT: extr w10, w15, w10, #2
+; CHECK-NEXT: asr w13, w13, #24
+; CHECK-NEXT: cmp w15, #1
+; CHECK-NEXT: csel w10, w8, w10, gt
+; CHECK-NEXT: cmn w15, #2
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: mov v2.b[12], w13
+; CHECK-NEXT: lsr x13, x11, #32
+; CHECK-NEXT: csel w10, w9, w10, lt
+; CHECK-NEXT: smull x12, w14, w12
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: extr w11, w13, w11, #2
+; CHECK-NEXT: cmp w13, #1
+; CHECK-NEXT: mov v2.b[13], w10
+; CHECK-NEXT: csel w10, w8, w11, gt
+; CHECK-NEXT: cmn w13, #2
+; CHECK-NEXT: lsr x11, x12, #32
+; CHECK-NEXT: csel w10, w9, w10, lt
+; CHECK-NEXT: asr w10, w10, #24
+; CHECK-NEXT: extr w12, w11, w12, #2
+; CHECK-NEXT: cmp w11, #1
+; CHECK-NEXT: mov v2.b[14], w10
+; CHECK-NEXT: csel w8, w8, w12, gt
+; CHECK-NEXT: cmn w11, #2
+; CHECK-NEXT: csel w8, w9, w8, lt
+; CHECK-NEXT: asr w8, w8, #24
+; CHECK-NEXT: mov v2.b[15], w8
+; CHECK-NEXT: mov v0.16b, v2.16b
+; CHECK-NEXT: ret
+ %tmp = call <16 x i8> @llvm.smul.fix.sat.v16i8(<16 x i8> %x, <16 x i8> %y, i32 2)
+ ret <16 x i8> %tmp
+}
+
+define <4 x i16> @vec_v4i16(<4 x i16> %x, <4 x i16> %y) {
+; CHECK-LABEL: vec_v4i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: umov w9, v0.h[0]
+; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NEXT: smov x10, v1.h[0]
+; CHECK-NEXT: mov w8, #2147483647 // =0x7fffffff
+; CHECK-NEXT: umov w11, v0.h[1]
+; CHECK-NEXT: smov x12, v1.h[1]
+; CHECK-NEXT: smov x13, v1.h[2]
+; CHECK-NEXT: umov w14, v0.h[3]
+; CHECK-NEXT: lsl w9, w9, #16
+; CHECK-NEXT: lsl w11, w11, #16
+; CHECK-NEXT: smull x9, w10, w9
+; CHECK-NEXT: umov w10, v0.h[2]
+; CHECK-NEXT: lsl w14, w14, #16
+; CHECK-NEXT: smull x11, w12, w11
+; CHECK-NEXT: lsr x12, x9, #32
+; CHECK-NEXT: lsl w10, w10, #16
+; CHECK-NEXT: lsr x15, x11, #32
+; CHECK-NEXT: extr w9, w12, w9, #10
+; CHECK-NEXT: cmp w12, #511
+; CHECK-NEXT: smull x10, w13, w10
+; CHECK-NEXT: mov w13, #-2147483648 // =0x80000000
+; CHECK-NEXT: extr w11, w15, w11, #10
+; CHECK-NEXT: csel w9, w8, w9, gt
+; CHECK-NEXT: cmn w12, #512
+; CHECK-NEXT: smov x12, v1.h[3]
+; CHECK-NEXT: csel w9, w13, w9, lt
+; CHECK-NEXT: cmp w15, #511
+; CHECK-NEXT: asr w9, w9, #16
+; CHECK-NEXT: csel w11, w8, w11, gt
+; CHECK-NEXT: cmn w15, #512
+; CHECK-NEXT: lsr x15, x10, #32
+; CHECK-NEXT: csel w11, w13, w11, lt
+; CHECK-NEXT: smull x12, w12, w14
+; CHECK-NEXT: fmov s0, w9
+; CHECK-NEXT: asr w11, w11, #16
+; CHECK-NEXT: extr w9, w15, w10, #10
+; CHECK-NEXT: cmp w15, #511
+; CHECK-NEXT: mov v0.h[1], w11
+; CHECK-NEXT: csel w9, w8, w9, gt
+; CHECK-NEXT: cmn w15, #512
+; CHECK-NEXT: lsr x10, x12, #32
+; CHECK-NEXT: csel w9, w13, w9, lt
+; CHECK-NEXT: asr w9, w9, #16
+; CHECK-NEXT: extr w11, w10, w12, #10
+; CHECK-NEXT: cmp w10, #511
+; CHECK-NEXT: mov v0.h[2], w9
+; CHECK-NEXT: csel w8, w8, w11, gt
+; CHECK-NEXT: cmn w10, #512
+; CHECK-NEXT: csel w8, w13, w8, lt
+; CHECK-NEXT: asr w8, w8, #16
+; CHECK-NEXT: mov v0.h[3], w8
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT: ret
+ %tmp = call <4 x i16> @llvm.smul.fix.sat.v4i16(<4 x i16> %x, <4 x i16> %y, i32 10)
+ ret <4 x i16> %tmp
+}
+
+define <8 x i16> @vec_v8i16(<8 x i16> %x, <8 x i16> %y) {
+; CHECK-LABEL: vec_v8i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: umov w8, v0.h[1]
+; CHECK-NEXT: smov x9, v1.h[1]
+; CHECK-NEXT: umov w10, v0.h[0]
+; CHECK-NEXT: smov x11, v1.h[0]
+; CHECK-NEXT: umov w13, v0.h[2]
+; CHECK-NEXT: umov w15, v0.h[3]
+; CHECK-NEXT: smov x16, v1.h[3]
+; CHECK-NEXT: lsl w8, w8, #16
+; CHECK-NEXT: lsl w13, w13, #16
+; CHECK-NEXT: lsl w15, w15, #16
+; CHECK-NEXT: smull x12, w9, w8
+; CHECK-NEXT: lsl w8, w10, #16
+; CHECK-NEXT: smov x10, v1.h[2]
+; CHECK-NEXT: mov w9, #2147483647 // =0x7fffffff
+; CHECK-NEXT: smull x15, w16, w15
+; CHECK-NEXT: umov w16, v0.h[5]
+; CHECK-NEXT: smull x11, w11, w8
+; CHECK-NEXT: mov w8, #-2147483648 // =0x80000000
+; CHECK-NEXT: lsr x14, x12, #32
+; CHECK-NEXT: smull x10, w10, w13
+; CHECK-NEXT: umov w13, v0.h[4]
+; CHECK-NEXT: extr w12, w14, w12, #2
+; CHECK-NEXT: lsr x17, x11, #32
+; CHECK-NEXT: cmp w14, #1
+; CHECK-NEXT: csel w12, w9, w12, gt
+; CHECK-NEXT: cmn w14, #2
+; CHECK-NEXT: extr w11, w17, w11, #2
+; CHECK-NEXT: csel w12, w8, w12, lt
+; CHECK-NEXT: cmp w17, #1
+; CHECK-NEXT: smov x14, v1.h[4]
+; CHECK-NEXT: csel w11, w9, w11, gt
+; CHECK-NEXT: cmn w17, #2
+; CHECK-NEXT: lsr x17, x10, #32
+; CHECK-NEXT: csel w11, w8, w11, lt
+; CHECK-NEXT: lsl w13, w13, #16
+; CHECK-NEXT: asr w12, w12, #16
+; CHECK-NEXT: asr w11, w11, #16
+; CHECK-NEXT: extr w10, w17, w10, #2
+; CHECK-NEXT: cmp w17, #1
+; CHECK-NEXT: smull x13, w14, w13
+; CHECK-NEXT: lsr x14, x15, #32
+; CHECK-NEXT: fmov s2, w11
+; CHECK-NEXT: smov x11, v1.h[5]
+; CHECK-NEXT: csel w10, w9, w10, gt
+; CHECK-NEXT: cmn w17, #2
+; CHECK-NEXT: extr w15, w14, w15, #2
+; CHECK-NEXT: csel w10, w8, w10, lt
+; CHECK-NEXT: cmp w14, #1
+; CHECK-NEXT: mov v2.h[1], w12
+; CHECK-NEXT: lsl w12, w16, #16
+; CHECK-NEXT: asr w10, w10, #16
+; CHECK-NEXT: umov w16, v0.h[6]
+; CHECK-NEXT: csel w15, w9, w15, gt
+; CHECK-NEXT: cmn w14, #2
+; CHECK-NEXT: smull x11, w11, w12
+; CHECK-NEXT: lsr x12, x13, #32
+; CHECK-NEXT: csel w14, w8, w15, lt
+; CHECK-NEXT: asr w14, w14, #16
+; CHECK-NEXT: mov v2.h[2], w10
+; CHECK-NEXT: smov x10, v1.h[6]
+; CHECK-NEXT: extr w13, w12, w13, #2
+; CHECK-NEXT: cmp w12, #1
+; CHECK-NEXT: lsl w15, w16, #16
+; CHECK-NEXT: lsr x16, x11, #32
+; CHECK-NEXT: csel w13, w9, w13, gt
+; CHECK-NEXT: cmn w12, #2
+; CHECK-NEXT: umov w12, v0.h[7]
+; CHECK-NEXT: csel w13, w8, w13, lt
+; CHECK-NEXT: extr w11, w16, w11, #2
+; CHECK-NEXT: cmp w16, #1
+; CHECK-NEXT: mov v2.h[3], w14
+; CHECK-NEXT: smull x10, w10, w15
+; CHECK-NEXT: smov x14, v1.h[7]
+; CHECK-NEXT: asr w13, w13, #16
+; CHECK-NEXT: csel w11, w9, w11, gt
+; CHECK-NEXT: cmn w16, #2
+; CHECK-NEXT: lsl w12, w12, #16
+; CHECK-NEXT: csel w11, w8, w11, lt
+; CHECK-NEXT: asr w11, w11, #16
+; CHECK-NEXT: mov v2.h[4], w13
+; CHECK-NEXT: lsr x13, x10, #32
+; CHECK-NEXT: smull x12, w14, w12
+; CHECK-NEXT: extr w10, w13, w10, #2
+; CHECK-NEXT: cmp w13, #1
+; CHECK-NEXT: mov v2.h[5], w11
+; CHECK-NEXT: csel w10, w9, w10, gt
+; CHECK-NEXT: cmn w13, #2
+; CHECK-NEXT: lsr x11, x12, #32
+; CHECK-NEXT: csel w10, w8, w10, lt
+; CHECK-NEXT: asr w10, w10, #16
+; CHECK-NEXT: extr w12, w11, w12, #2
+; CHECK-NEXT: cmp w11, #1
+; CHECK-NEXT: mov v2.h[6], w10
+; CHECK-NEXT: csel w9, w9, w12, gt
+; CHECK-NEXT: cmn w11, #2
+; CHECK-NEXT: csel w8, w8, w9, lt
+; CHECK-NEXT: asr w8, w8, #16
+; CHECK-NEXT: mov v2.h[7], w8
+; CHECK-NEXT: mov v0.16b, v2.16b
+; CHECK-NEXT: ret
+ %tmp = call <8 x i16> @llvm.smul.fix.sat.v8i16(<8 x i16> %x, <8 x i16> %y, i32 2)
+ ret <8 x i16> %tmp
+}
+
+define <2 x i32> @vec_v2i32(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: vec_v2i32:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
@@ -173,53 +672,169 @@ define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
ret <2 x i32> %tmp
}
-define <4 x i32> @vec2(<4 x i32> %x, <4 x i32> %y) nounwind {
-; CHECK-LABEL: vec2:
+define <4 x i32> @vec_v4i32(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: vec_v4i32:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w9, v1.s[1]
; CHECK-NEXT: mov w10, v0.s[1]
-; CHECK-NEXT: mov w8, #-2147483648 // =0x80000000
-; CHECK-NEXT: fmov w12, s1
-; CHECK-NEXT: fmov w13, s0
-; CHECK-NEXT: mov w14, v0.s[2]
-; CHECK-NEXT: eor w11, w10, w9
+; CHECK-NEXT: mov w8, #2147483647 // =0x7fffffff
+; CHECK-NEXT: fmov w11, s0
+; CHECK-NEXT: mov w12, v1.s[2]
+; CHECK-NEXT: mov w13, v0.s[2]
+; CHECK-NEXT: mov w14, v1.s[3]
+; CHECK-NEXT: mov w16, v0.s[3]
; CHECK-NEXT: smull x9, w10, w9
-; CHECK-NEXT: mov w10, v1.s[2]
-; CHECK-NEXT: cmp w11, #0
+; CHECK-NEXT: fmov w10, s1
+; CHECK-NEXT: smull x10, w11, w10
+; CHECK-NEXT: lsr x11, x9, #32
+; CHECK-NEXT: extr w9, w11, w9, #15
+; CHECK-NEXT: cmp w11, #4, lsl #12 // =16384
+; CHECK-NEXT: lsr x15, x10, #32
+; CHECK-NEXT: csel w9, w8, w9, ge
+; CHECK-NEXT: cmn w11, #4, lsl #12 // =16384
; CHECK-NEXT: smull x11, w13, w12
-; CHECK-NEXT: eor w12, w13, w12
+; CHECK-NEXT: mov w12, #-2147483648 // =0x80000000
+; CHECK-NEXT: extr w10, w15, w10, #15
+; CHECK-NEXT: csel w9, w12, w9, lt
+; CHECK-NEXT: cmp w15, #4, lsl #12 // =16384
+; CHECK-NEXT: csel w10, w8, w10, ge
+; CHECK-NEXT: cmn w15, #4, lsl #12 // =16384
+; CHECK-NEXT: lsr x13, x11, #32
+; CHECK-NEXT: csel w10, w12, w10, lt
+; CHECK-NEXT: fmov s0, w10
+; CHECK-NEXT: smull x10, w16, w14
+; CHECK-NEXT: cmp w13, #4, lsl #12 // =16384
+; CHECK-NEXT: mov v0.s[1], w9
+; CHECK-NEXT: extr w9, w13, w11, #15
+; CHECK-NEXT: lsr x11, x10, #32
+; CHECK-NEXT: csel w9, w8, w9, ge
+; CHECK-NEXT: cmn w13, #4, lsl #12 // =16384
+; CHECK-NEXT: csel w9, w12, w9, lt
+; CHECK-NEXT: cmp w11, #4, lsl #12 // =16384
+; CHECK-NEXT: mov v0.s[2], w9
+; CHECK-NEXT: extr w9, w11, w10, #15
+; CHECK-NEXT: csel w8, w8, w9, ge
+; CHECK-NEXT: cmn w11, #4, lsl #12 // =16384
+; CHECK-NEXT: csel w8, w12, w8, lt
+; CHECK-NEXT: mov v0.s[3], w8
+; CHECK-NEXT: ret
+ %tmp = call <4 x i32> @llvm.smul.fix.sat.v4i32(<4 x i32> %x, <4 x i32> %y, i32 15)
+ ret <4 x i32> %tmp
+}
+
+define <8 x i32> @vec_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; CHECK-LABEL: vec_v8i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov w9, v2.s[1]
+; CHECK-NEXT: mov w10, v0.s[1]
+; CHECK-NEXT: mov w8, #-2147483648 // =0x80000000
+; CHECK-NEXT: fmov w13, s2
+; CHECK-NEXT: fmov w16, s0
+; CHECK-NEXT: mov w11, v2.s[2]
+; CHECK-NEXT: mov w14, v0.s[2]
+; CHECK-NEXT: mov w12, v2.s[3]
+; CHECK-NEXT: mov w15, v0.s[3]
+; CHECK-NEXT: mov w18, v3.s[1]
+; CHECK-NEXT: mov w0, v1.s[1]
+; CHECK-NEXT: smull x17, w10, w9
+; CHECK-NEXT: eor w9, w10, w9
+; CHECK-NEXT: mov w10, v3.s[2]
+; CHECK-NEXT: cmp w9, #0
+; CHECK-NEXT: smull x9, w16, w13
+; CHECK-NEXT: eor w13, w16, w13
+; CHECK-NEXT: cinv w16, w8, pl
+; CHECK-NEXT: cmp x17, w17, sxtw
+; CHECK-NEXT: csel w16, w16, w17, ne
+; CHECK-NEXT: smull x17, w14, w11
+; CHECK-NEXT: cmp w13, #0
+; CHECK-NEXT: eor w11, w14, w11
; CHECK-NEXT: cinv w13, w8, pl
; CHECK-NEXT: cmp x9, w9, sxtw
; CHECK-NEXT: csel w9, w13, w9, ne
-; CHECK-NEXT: cmp w12, #0
-; CHECK-NEXT: mov w13, v1.s[3]
+; CHECK-NEXT: cmp w11, #0
+; CHECK-NEXT: smull x14, w15, w12
+; CHECK-NEXT: eor w11, w15, w12
; CHECK-NEXT: cinv w12, w8, pl
+; CHECK-NEXT: cmp x17, w17, sxtw
+; CHECK-NEXT: eor w13, w0, w18
+; CHECK-NEXT: csel w12, w12, w17, ne
+; CHECK-NEXT: cmp w11, #0
+; CHECK-NEXT: smull x11, w0, w18
+; CHECK-NEXT: fmov w17, s3
+; CHECK-NEXT: fmov w0, s1
+; CHECK-NEXT: cinv w15, w8, pl
+; CHECK-NEXT: cmp x14, w14, sxtw
+; CHECK-NEXT: mov w18, v1.s[2]
+; CHECK-NEXT: csel w14, w15, w14, ne
+; CHECK-NEXT: cmp w13, #0
+; CHECK-NEXT: fmov s0, w9
+; CHECK-NEXT: smull x13, w0, w17
+; CHECK-NEXT: cinv w15, w8, pl
; CHECK-NEXT: cmp x11, w11, sxtw
-; CHECK-NEXT: csel w11, w12, w11, ne
-; CHECK-NEXT: mov w12, v0.s[3]
-; CHECK-NEXT: fmov s0, w11
-; CHECK-NEXT: smull x11, w14, w10
-; CHECK-NEXT: mov v0.s[1], w9
-; CHECK-NEXT: eor w9, w14, w10
-; CHECK-NEXT: smull x10, w12, w13
-; CHECK-NEXT: cmp w9, #0
-; CHECK-NEXT: cinv w9, w8, pl
-; CHECK-NEXT: cmp x11, w11, sxtw
-; CHECK-NEXT: csel w9, w9, w11, ne
-; CHECK-NEXT: mov v0.s[2], w9
-; CHECK-NEXT: eor w9, w12, w13
+; CHECK-NEXT: eor w17, w0, w17
+; CHECK-NEXT: csel w11, w15, w11, ne
+; CHECK-NEXT: mov w0, v1.s[3]
+; CHECK-NEXT: cmp w17, #0
+; CHECK-NEXT: mov w17, v3.s[3]
+; CHECK-NEXT: smull x9, w18, w10
+; CHECK-NEXT: cinv w15, w8, pl
+; CHECK-NEXT: cmp x13, w13, sxtw
+; CHECK-NEXT: mov v0.s[1], w16
+; CHECK-NEXT: csel w13, w15, w13, ne
+; CHECK-NEXT: eor w10, w18, w10
+; CHECK-NEXT: fmov s1, w13
+; CHECK-NEXT: cmp w10, #0
+; CHECK-NEXT: cinv w10, w8, pl
+; CHECK-NEXT: cmp x9, w9, sxtw
+; CHECK-NEXT: csel w9, w10, w9, ne
+; CHECK-NEXT: smull x10, w0, w17
+; CHECK-NEXT: mov v0.s[2], w12
+; CHECK-NEXT: mov v1.s[1], w11
+; CHECK-NEXT: mov v0.s[3], w14
+; CHECK-NEXT: mov v1.s[2], w9
+; CHECK-NEXT: eor w9, w0, w17
; CHECK-NEXT: cmp w9, #0
; CHECK-NEXT: cinv w8, w8, pl
; CHECK-NEXT: cmp x10, w10, sxtw
; CHECK-NEXT: csel w8, w8, w10, ne
-; CHECK-NEXT: mov v0.s[3], w8
+; CHECK-NEXT: mov v1.s[3], w8
; CHECK-NEXT: ret
- %tmp = call <4 x i32> @llvm.smul.fix.sat.v4i32(<4 x i32> %x, <4 x i32> %y, i32 0)
- ret <4 x i32> %tmp
+ %tmp = call <8 x i32> @llvm.smul.fix.sat.v8i32(<8 x i32> %x, <8 x i32> %y, i32 0)
+ ret <8 x i32> %tmp
+}
+
+define <2 x i64> @vec_v2i64(<2 x i64> %x, <2 x i64> %y) {
+; CHECK-LABEL: vec_v2i64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov x9, v1.d[1]
+; CHECK-NEXT: mov x10, v0.d[1]
+; CHECK-NEXT: mov x8, #9223372036854775807 // =0x7fffffffffffffff
+; CHECK-NEXT: fmov x12, d0
+; CHECK-NEXT: mul x11, x10, x9
+; CHECK-NEXT: smulh x9, x10, x9
+; CHECK-NEXT: fmov x10, d1
+; CHECK-NEXT: mul x13, x12, x10
+; CHECK-NEXT: smulh x10, x12, x10
+; CHECK-NEXT: extr x11, x9, x11, #15
+; CHECK-NEXT: cmp x9, #4, lsl #12 // =16384
+; CHECK-NEXT: csel x11, x8, x11, ge
+; CHECK-NEXT: cmn x9, #4, lsl #12 // =16384
+; CHECK-NEXT: mov x9, #-9223372036854775808 // =0x8000000000000000
+; CHECK-NEXT: csel x11, x9, x11, lt
+; CHECK-NEXT: extr x12, x10, x13, #15
+; CHECK-NEXT: cmp x10, #4, lsl #12 // =16384
+; CHECK-NEXT: csel x8, x8, x12, ge
+; CHECK-NEXT: cmn x10, #4, lsl #12 // =16384
+; CHECK-NEXT: csel x8, x9, x8, lt
+; CHECK-NEXT: fmov d0, x8
+; CHECK-NEXT: mov v0.d[1], x11
+; CHECK-NEXT: ret
+ %tmp = call <2 x i64> @llvm.smul.fix.sat.v2i64(<2 x i64> %x, <2 x i64> %y, i32 15)
+ ret <2 x i64> %tmp
}
-define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
-; CHECK-LABEL: vec3:
+define <4 x i64> @vec_v4i64(<4 x i64> %x, <4 x i64> %y) {
+; CHECK-LABEL: vec_v4i64:
; CHECK: // %bb.0:
; CHECK-NEXT: mov x8, v2.d[1]
; CHECK-NEXT: mov x9, v0.d[1]
@@ -269,3 +884,7 @@ define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
%tmp = call <4 x i64> @llvm.smul.fix.sat.v4i64(<4 x i64> %x, <4 x i64> %y, i32 32)
ret <4 x i64> %tmp
}
+
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK-GI: {{.*}}
+; CHECK-SD: {{.*}}
diff --git a/llvm/test/CodeGen/AArch64/umul_fix.ll b/llvm/test/CodeGen/AArch64/umul_fix.ll
index 0656f5e17b30e..32483ff2da4d6 100644
--- a/llvm/test/CodeGen/AArch64/umul_fix.ll
+++ b/llvm/test/CodeGen/AArch64/umul_fix.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=0 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
-; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-define i32 @func(i32 %x, i32 %y) nounwind {
+define i32 @func(i32 %x, i32 %y) {
; CHECK-SD-LABEL: func:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: umull x8, w0, w1
@@ -20,7 +20,7 @@ define i32 @func(i32 %x, i32 %y) nounwind {
ret i32 %tmp
}
-define i64 @func2(i64 %x, i64 %y) nounwind {
+define i64 @func2(i64 %x, i64 %y) {
; CHECK-LABEL: func2:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x8, x0, x1
@@ -31,7 +31,7 @@ define i64 @func2(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i4 @func3(i4 %x, i4 %y) nounwind {
+define i4 @func3(i4 %x, i4 %y) {
; CHECK-SD-LABEL: func3:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: and w8, w1, #0xf
@@ -52,7 +52,7 @@ define i4 @func3(i4 %x, i4 %y) nounwind {
}
;; These result in regular integer multiplication
-define i32 @func4(i32 %x, i32 %y) nounwind {
+define i32 @func4(i32 %x, i32 %y) {
; CHECK-LABEL: func4:
; CHECK: // %bb.0:
; CHECK-NEXT: mul w0, w0, w1
@@ -61,7 +61,7 @@ define i32 @func4(i32 %x, i32 %y) nounwind {
ret i32 %tmp
}
-define i64 @func5(i64 %x, i64 %y) nounwind {
+define i64 @func5(i64 %x, i64 %y) {
; CHECK-LABEL: func5:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x0, x0, x1
@@ -70,7 +70,7 @@ define i64 @func5(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i4 @func6(i4 %x, i4 %y) nounwind {
+define i4 @func6(i4 %x, i4 %y) {
; CHECK-SD-LABEL: func6:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: and w8, w1, #0xf
@@ -86,7 +86,7 @@ define i4 @func6(i4 %x, i4 %y) nounwind {
ret i4 %tmp
}
-define i64 @func7(i64 %x, i64 %y) nounwind {
+define i64 @func7(i64 %x, i64 %y) {
; CHECK-LABEL: func7:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x8, x0, x1
@@ -97,7 +97,7 @@ define i64 @func7(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i64 @func8(i64 %x, i64 %y) nounwind {
+define i64 @func8(i64 %x, i64 %y) {
; CHECK-LABEL: func8:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x8, x0, x1
@@ -108,7 +108,7 @@ define i64 @func8(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i64 @func9(i64 %x, i64 %y) nounwind {
+define i64 @func9(i64 %x, i64 %y) {
; CHECK-LABEL: func9:
; CHECK: // %bb.0:
; CHECK-NEXT: umulh x0, x0, x1
@@ -117,8 +117,90 @@ define i64 @func9(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
-; CHECK-LABEL: vec:
+define <8 x i8> @vec_v8i8(<8 x i8> %x, <8 x i8> %y) {
+; CHECK-SD-LABEL: vec_v8i8:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: umull v0.8h, v0.8b, v1.8b
+; CHECK-SD-NEXT: shrn v1.8b, v0.8h, #8
+; CHECK-SD-NEXT: xtn v2.8b, v0.8h
+; CHECK-SD-NEXT: shl v0.8b, v1.8b, #6
+; CHECK-SD-NEXT: usra v0.8b, v2.8b, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i8:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: umull v0.8h, v0.8b, v1.8b
+; CHECK-GI-NEXT: shrn v0.8b, v0.8h, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i8> @llvm.umul.fix.v8i8(<8 x i8> %x, <8 x i8> %y, i32 2)
+ ret <8 x i8> %tmp
+}
+
+define <16 x i8> @vec_v16i8(<16 x i8> %x, <16 x i8> %y) {
+; CHECK-SD-LABEL: vec_v16i8:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: umull2 v2.8h, v0.16b, v1.16b
+; CHECK-SD-NEXT: umull v3.8h, v0.8b, v1.8b
+; CHECK-SD-NEXT: mul v1.16b, v0.16b, v1.16b
+; CHECK-SD-NEXT: uzp2 v2.16b, v3.16b, v2.16b
+; CHECK-SD-NEXT: shl v0.16b, v2.16b, #6
+; CHECK-SD-NEXT: usra v0.16b, v1.16b, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v16i8:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: umull v2.8h, v0.8b, v1.8b
+; CHECK-GI-NEXT: umull2 v1.8h, v0.16b, v1.16b
+; CHECK-GI-NEXT: shrn v0.8b, v2.8h, #2
+; CHECK-GI-NEXT: shrn2 v0.16b, v1.8h, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> %x, <16 x i8> %y, i32 2)
+ ret <16 x i8> %tmp
+}
+
+define <4 x i16> @vec_v4i16(<4 x i16> %x, <4 x i16> %y) {
+; CHECK-SD-LABEL: vec_v4i16:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: umull v0.4s, v0.4h, v1.4h
+; CHECK-SD-NEXT: shrn v1.4h, v0.4s, #16
+; CHECK-SD-NEXT: xtn v2.4h, v0.4s
+; CHECK-SD-NEXT: shl v0.4h, v1.4h, #6
+; CHECK-SD-NEXT: usra v0.4h, v2.4h, #10
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v4i16:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: umull v0.4s, v0.4h, v1.4h
+; CHECK-GI-NEXT: shrn v0.4h, v0.4s, #10
+; CHECK-GI-NEXT: ret
+ %tmp = call <4 x i16> @llvm.umul.fix.v4i16(<4 x i16> %x, <4 x i16> %y, i32 10)
+ ret <4 x i16> %tmp
+}
+
+define <8 x i16> @vec_v8i16(<8 x i16> %x, <8 x i16> %y) {
+; CHECK-SD-LABEL: vec_v8i16:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: umull2 v2.4s, v0.8h, v1.8h
+; CHECK-SD-NEXT: umull v3.4s, v0.4h, v1.4h
+; CHECK-SD-NEXT: mul v1.8h, v0.8h, v1.8h
+; CHECK-SD-NEXT: uzp2 v2.8h, v3.8h, v2.8h
+; CHECK-SD-NEXT: shl v0.8h, v2.8h, #14
+; CHECK-SD-NEXT: usra v0.8h, v1.8h, #2
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i16:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: umull v2.4s, v0.4h, v1.4h
+; CHECK-GI-NEXT: umull2 v1.4s, v0.8h, v1.8h
+; CHECK-GI-NEXT: shrn v0.4h, v2.4s, #2
+; CHECK-GI-NEXT: shrn2 v0.8h, v1.4s, #2
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> %x, <8 x i16> %y, i32 2)
+ ret <8 x i16> %tmp
+}
+
+define <2 x i32> @vec_v2i32(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: vec_v2i32:
; CHECK: // %bb.0:
; CHECK-NEXT: mul v0.2s, v0.2s, v1.2s
; CHECK-NEXT: ret
@@ -126,17 +208,84 @@ define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
ret <2 x i32> %tmp
}
-define <4 x i32> @vec2(<4 x i32> %x, <4 x i32> %y) nounwind {
-; CHECK-LABEL: vec2:
-; CHECK: // %bb.0:
-; CHECK-NEXT: mul v0.4s, v0.4s, v1.4s
-; CHECK-NEXT: ret
- %tmp = call <4 x i32> @llvm.umul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 0)
+define <4 x i32> @vec_v4i32(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-SD-LABEL: vec_v4i32:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: umull2 v2.2d, v0.4s, v1.4s
+; CHECK-SD-NEXT: umull v3.2d, v0.2s, v1.2s
+; CHECK-SD-NEXT: mul v1.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT: uzp2 v2.4s, v3.4s, v2.4s
+; CHECK-SD-NEXT: shl v0.4s, v2.4s, #17
+; CHECK-SD-NEXT: usra v0.4s, v1.4s, #15
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v4i32:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: umull v2.2d, v0.2s, v1.2s
+; CHECK-GI-NEXT: umull2 v1.2d, v0.4s, v1.4s
+; CHECK-GI-NEXT: shrn v0.2s, v2.2d, #15
+; CHECK-GI-NEXT: shrn2 v0.4s, v1.2d, #15
+; CHECK-GI-NEXT: ret
+ %tmp = call <4 x i32> @llvm.umul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 15)
ret <4 x i32> %tmp
}
-define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
-; CHECK-SD-LABEL: vec3:
+define <8 x i32> @vec_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; CHECK-SD-LABEL: vec_v8i32:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: mul v1.4s, v1.4s, v3.4s
+; CHECK-SD-NEXT: mul v0.4s, v0.4s, v2.4s
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v8i32:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: mul v0.4s, v0.4s, v2.4s
+; CHECK-GI-NEXT: mul v1.4s, v1.4s, v3.4s
+; CHECK-GI-NEXT: ret
+ %tmp = call <8 x i32> @llvm.umul.fix.v8i32(<8 x i32> %x, <8 x i32> %y, i32 0)
+ ret <8 x i32> %tmp
+}
+
+define <2 x i64> @vec_v2i64(<2 x i64> %x, <2 x i64> %y) {
+; CHECK-SD-LABEL: vec_v2i64:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: fmov x8, d1
+; CHECK-SD-NEXT: fmov x9, d0
+; CHECK-SD-NEXT: mov x10, v1.d[1]
+; CHECK-SD-NEXT: mov x11, v0.d[1]
+; CHECK-SD-NEXT: mul x12, x9, x8
+; CHECK-SD-NEXT: umulh x8, x9, x8
+; CHECK-SD-NEXT: mul x9, x11, x10
+; CHECK-SD-NEXT: umulh x10, x11, x10
+; CHECK-SD-NEXT: extr x8, x8, x12, #15
+; CHECK-SD-NEXT: fmov d0, x8
+; CHECK-SD-NEXT: extr x9, x10, x9, #15
+; CHECK-SD-NEXT: mov v0.d[1], x9
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: vec_v2i64:
+; CHECK-GI: // %bb.0:
+; CHECK-GI-NEXT: mov d2, v0.d[1]
+; CHECK-GI-NEXT: mov d3, v1.d[1]
+; CHECK-GI-NEXT: fmov x8, d0
+; CHECK-GI-NEXT: fmov x9, d1
+; CHECK-GI-NEXT: mul x10, x8, x9
+; CHECK-GI-NEXT: fmov x11, d3
+; CHECK-GI-NEXT: umulh x8, x8, x9
+; CHECK-GI-NEXT: fmov x9, d2
+; CHECK-GI-NEXT: mul x12, x9, x11
+; CHECK-GI-NEXT: umulh x9, x9, x11
+; CHECK-GI-NEXT: extr x8, x8, x10, #15
+; CHECK-GI-NEXT: fmov d0, x8
+; CHECK-GI-NEXT: extr x9, x9, x12, #15
+; CHECK-GI-NEXT: mov v0.d[1], x9
+; CHECK-GI-NEXT: ret
+ %tmp = call <2 x i64> @llvm.umul.fix.v2i64(<2 x i64> %x, <2 x i64> %y, i32 15)
+ ret <2 x i64> %tmp
+}
+
+define <4 x i64> @vec_v4i64(<4 x i64> %x, <4 x i64> %y) {
+; CHECK-SD-LABEL: vec_v4i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov x8, v2.d[1]
; CHECK-SD-NEXT: mov x9, v0.d[1]
@@ -164,7 +313,7 @@ define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
; CHECK-SD-NEXT: mov v1.d[1], x11
; CHECK-SD-NEXT: ret
;
-; CHECK-GI-LABEL: vec3:
+; CHECK-GI-LABEL: vec_v4i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: mov d4, v0.d[1]
; CHECK-GI-NEXT: mov d5, v2.d[1]
@@ -199,7 +348,7 @@ define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
ret <4 x i64> %tmp
}
-define <4 x i16> @widemul(<4 x i16> %x, <4 x i16> %y) nounwind {
+define <4 x i16> @widemul(<4 x i16> %x, <4 x i16> %y) {
; CHECK-SD-LABEL: widemul:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: umull v0.4s, v0.4h, v1.4h
diff --git a/llvm/test/CodeGen/AArch64/umul_fix_sat.ll b/llvm/test/CodeGen/AArch64/umul_fix_sat.ll
index e9965fd5319b5..3b319ca5a9676 100644
--- a/llvm/test/CodeGen/AArch64/umul_fix_sat.ll
+++ b/llvm/test/CodeGen/AArch64/umul_fix_sat.ll
@@ -1,7 +1,26 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=aarch64-linux-gnu | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=0 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
+; RUN: llc < %s -mtriple=aarch64-linux-gnu -global-isel=1 -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-define i32 @func(i32 %x, i32 %y) nounwind {
+; CHECK-GI: warning: Instruction selection used fallback path for func
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func2
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func3
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func4
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func5
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func6
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func7
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for func8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v16i8
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i16
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i16
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v2i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v8i32
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v2i64
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for vec_v4i64
+
+define i32 @func(i32 %x, i32 %y) {
; CHECK-LABEL: func:
; CHECK: // %bb.0:
; CHECK-NEXT: umull x8, w0, w1
@@ -14,7 +33,7 @@ define i32 @func(i32 %x, i32 %y) nounwind {
ret i32 %tmp
}
-define i64 @func2(i64 %x, i64 %y) nounwind {
+define i64 @func2(i64 %x, i64 %y) {
; CHECK-LABEL: func2:
; CHECK: // %bb.0:
; CHECK-NEXT: mul x8, x0, x1
@@ -27,7 +46,7 @@ define i64 @func2(i64 %x, i64 %y) nounwind {
ret i64 %tmp
}
-define i4 @func3(i4 %x, i4 %y) nounwind {
+define i4 @func3(i4 %x, i4 %y) {
; CHECK-LABEL: func3:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl w8, w0, #28
@@ -44,7 +63,7 @@ define i4 @func3(i4 %x, i4 %y) nounwind {
}
;; These result in regular integer multiplication with a saturation check.
-define i32 @func4(i32 %x, i32 %y) nounwind {
+define i32 @func4(i32 %x, i32 %y) {
; CHECK-LABEL: func4:
; CHECK: // %bb.0:
; CHECK-NEXT: umull x8, w0, w1
@@ -67,7 +86,7 @@ define i64 @func5(i64 %x, i64 %y) {
ret i64 %tmp
}
-define i4 @func6(i4 %x, i4 %y) nounwind {
+define i4 @func6(i4 %x, i4 %y) {
; CHECK-LABEL: func6:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl w8, w0, #28
@@ -81,8 +100,436 @@ define i4 @func6(i4 %x, i4 %y) nounwind {
ret i4 %tmp
}
-define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
-; CHECK-LABEL: vec:
+define i64 @func7(i64 %x, i64 %y) {
+; CHECK-LABEL: func7:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mul x9, x0, x1
+; CHECK-NEXT: mov w8, #-1 // =0xffffffff
+; CHECK-NEXT: umulh x10, x0, x1
+; CHECK-NEXT: extr x9, x10, x9, #32
+; CHECK-NEXT: cmp x10, x8
+; CHECK-NEXT: csinv x0, x9, xzr, ls
+; CHECK-NEXT: ret
+ %tmp = call i64 @llvm.umul.fix.sat.i64(i64 %x, i64 %y, i32 32)
+ ret i64 %tmp
+}
+
+define i64 @func8(i64 %x, i64 %y) {
+; CHECK-LABEL: func8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mul x9, x0, x1
+; CHECK-NEXT: mov x8, #9223372036854775807 // =0x7fffffffffffffff
+; CHECK-NEXT: umulh x10, x0, x1
+; CHECK-NEXT: extr x9, x10, x9, #63
+; CHECK-NEXT: cmp x10, x8
+; CHECK-NEXT: csinv x0, x9, xzr, ls
+; CHECK-NEXT: ret
+ %tmp = call i64 @llvm.umul.fix.sat.i64(i64 %x, i64 %y, i32 63)
+ ret i64 %tmp
+}
+
+define <8 x i8> @vec_v8i8(<8 x i8> %x, <8 x i8> %y) {
+; CHECK-LABEL: vec_v8i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: umov w8, v0.b[0]
+; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NEXT: umov w10, v1.b[0]
+; CHECK-NEXT: umov w9, v0.b[1]
+; CHECK-NEXT: umov w11, v1.b[1]
+; CHECK-NEXT: umov w12, v0.b[2]
+; CHECK-NEXT: lsl w8, w8, #24
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: umull x8, w8, w10
+; CHECK-NEXT: umov w10, v1.b[2]
+; CHECK-NEXT: umull x9, w9, w11
+; CHECK-NEXT: umov w11, v0.b[3]
+; CHECK-NEXT: lsr x13, x8, #32
+; CHECK-NEXT: lsr x14, x9, #32
+; CHECK-NEXT: umull x10, w12, w10
+; CHECK-NEXT: umov w12, v1.b[3]
+; CHECK-NEXT: extr w8, w13, w8, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: umov w13, v0.b[4]
+; CHECK-NEXT: extr w9, w14, w9, #2
+; CHECK-NEXT: lsl w11, w11, #24
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: lsr x14, x10, #32
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: umull x11, w11, w12
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: extr w10, w14, w10, #2
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: fmov s2, w8
+; CHECK-NEXT: umov w8, v1.b[4]
+; CHECK-NEXT: umov w12, v0.b[5]
+; CHECK-NEXT: csinv w10, w10, wzr, ls
+; CHECK-NEXT: lsr w10, w10, #24
+; CHECK-NEXT: mov v2.b[1], w9
+; CHECK-NEXT: lsl w9, w13, #24
+; CHECK-NEXT: lsr x13, x11, #32
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: umull x8, w9, w8
+; CHECK-NEXT: umov w9, v1.b[5]
+; CHECK-NEXT: extr w11, w13, w11, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.b[2], w10
+; CHECK-NEXT: umov w10, v0.b[6]
+; CHECK-NEXT: csinv w11, w11, wzr, ls
+; CHECK-NEXT: lsr x13, x8, #32
+; CHECK-NEXT: lsr w11, w11, #24
+; CHECK-NEXT: umull x9, w12, w9
+; CHECK-NEXT: umov w12, v1.b[6]
+; CHECK-NEXT: extr w8, w13, w8, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.b[3], w11
+; CHECK-NEXT: lsl w10, w10, #24
+; CHECK-NEXT: umov w11, v0.b[7]
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr x13, x9, #32
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: umull x10, w10, w12
+; CHECK-NEXT: umov w12, v1.b[7]
+; CHECK-NEXT: extr w9, w13, w9, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.b[4], w8
+; CHECK-NEXT: lsl w8, w11, #24
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: lsr x11, x10, #32
+; CHECK-NEXT: umull x8, w8, w12
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: extr w10, w11, w10, #2
+; CHECK-NEXT: cmp w11, #3
+; CHECK-NEXT: mov v2.b[5], w9
+; CHECK-NEXT: csinv w9, w10, wzr, ls
+; CHECK-NEXT: lsr x10, x8, #32
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: extr w8, w10, w8, #2
+; CHECK-NEXT: cmp w10, #3
+; CHECK-NEXT: mov v2.b[6], w9
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: mov v2.b[7], w8
+; CHECK-NEXT: fmov d0, d2
+; CHECK-NEXT: ret
+ %tmp = call <8 x i8> @llvm.umul.fix.sat.v8i8(<8 x i8> %x, <8 x i8> %y, i32 2)
+ ret <8 x i8> %tmp
+}
+
+define <16 x i8> @vec_v16i8(<16 x i8> %x, <16 x i8> %y) {
+; CHECK-LABEL: vec_v16i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: umov w8, v0.b[1]
+; CHECK-NEXT: umov w9, v0.b[0]
+; CHECK-NEXT: umov w10, v1.b[1]
+; CHECK-NEXT: umov w11, v1.b[0]
+; CHECK-NEXT: umov w12, v0.b[2]
+; CHECK-NEXT: umov w13, v0.b[3]
+; CHECK-NEXT: lsl w8, w8, #24
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: lsl w13, w13, #24
+; CHECK-NEXT: umull x8, w8, w10
+; CHECK-NEXT: umov w10, v1.b[2]
+; CHECK-NEXT: umull x9, w9, w11
+; CHECK-NEXT: lsl w11, w12, #24
+; CHECK-NEXT: lsr x12, x8, #32
+; CHECK-NEXT: lsr x14, x9, #32
+; CHECK-NEXT: umull x10, w11, w10
+; CHECK-NEXT: umov w11, v1.b[3]
+; CHECK-NEXT: extr w8, w12, w8, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: umov w12, v0.b[4]
+; CHECK-NEXT: extr w9, w14, w9, #2
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: lsr x14, x10, #32
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: umull x11, w13, w11
+; CHECK-NEXT: umov w13, v1.b[4]
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: extr w10, w14, w10, #2
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: fmov s2, w9
+; CHECK-NEXT: umov w9, v0.b[5]
+; CHECK-NEXT: csinv w10, w10, wzr, ls
+; CHECK-NEXT: umull x12, w12, w13
+; CHECK-NEXT: umov w13, v1.b[5]
+; CHECK-NEXT: lsr w10, w10, #24
+; CHECK-NEXT: mov v2.b[1], w8
+; CHECK-NEXT: lsr x8, x11, #32
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: extr w11, w8, w11, #2
+; CHECK-NEXT: cmp w8, #3
+; CHECK-NEXT: umov w8, v0.b[6]
+; CHECK-NEXT: umull x9, w9, w13
+; CHECK-NEXT: umov w13, v1.b[6]
+; CHECK-NEXT: mov v2.b[2], w10
+; CHECK-NEXT: lsr x10, x12, #32
+; CHECK-NEXT: csinv w11, w11, wzr, ls
+; CHECK-NEXT: lsr w11, w11, #24
+; CHECK-NEXT: extr w12, w10, w12, #2
+; CHECK-NEXT: cmp w10, #3
+; CHECK-NEXT: umov w10, v0.b[7]
+; CHECK-NEXT: lsl w8, w8, #24
+; CHECK-NEXT: mov v2.b[3], w11
+; CHECK-NEXT: lsr x11, x9, #32
+; CHECK-NEXT: csinv w12, w12, wzr, ls
+; CHECK-NEXT: umull x8, w8, w13
+; CHECK-NEXT: umov w13, v1.b[7]
+; CHECK-NEXT: lsr w12, w12, #24
+; CHECK-NEXT: extr w9, w11, w9, #2
+; CHECK-NEXT: cmp w11, #3
+; CHECK-NEXT: umov w11, v0.b[8]
+; CHECK-NEXT: lsl w10, w10, #24
+; CHECK-NEXT: mov v2.b[4], w12
+; CHECK-NEXT: lsr x12, x8, #32
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: umull x10, w10, w13
+; CHECK-NEXT: umov w13, v1.b[8]
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: extr w8, w12, w8, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: umov w12, v0.b[9]
+; CHECK-NEXT: lsl w11, w11, #24
+; CHECK-NEXT: mov v2.b[5], w9
+; CHECK-NEXT: lsr x9, x10, #32
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: umull x11, w11, w13
+; CHECK-NEXT: umov w13, v1.b[9]
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: extr w10, w9, w10, #2
+; CHECK-NEXT: cmp w9, #3
+; CHECK-NEXT: umov w9, v0.b[10]
+; CHECK-NEXT: lsl w12, w12, #24
+; CHECK-NEXT: mov v2.b[6], w8
+; CHECK-NEXT: lsr x8, x11, #32
+; CHECK-NEXT: csinv w10, w10, wzr, ls
+; CHECK-NEXT: umull x12, w12, w13
+; CHECK-NEXT: umov w13, v1.b[10]
+; CHECK-NEXT: lsr w10, w10, #24
+; CHECK-NEXT: extr w11, w8, w11, #2
+; CHECK-NEXT: cmp w8, #3
+; CHECK-NEXT: umov w8, v0.b[11]
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: mov v2.b[7], w10
+; CHECK-NEXT: lsr x10, x12, #32
+; CHECK-NEXT: csinv w11, w11, wzr, ls
+; CHECK-NEXT: umull x9, w9, w13
+; CHECK-NEXT: umov w13, v1.b[11]
+; CHECK-NEXT: lsr w11, w11, #24
+; CHECK-NEXT: extr w12, w10, w12, #2
+; CHECK-NEXT: lsl w8, w8, #24
+; CHECK-NEXT: cmp w10, #3
+; CHECK-NEXT: umov w10, v0.b[12]
+; CHECK-NEXT: mov v2.b[8], w11
+; CHECK-NEXT: lsr x11, x9, #32
+; CHECK-NEXT: csinv w12, w12, wzr, ls
+; CHECK-NEXT: umull x8, w8, w13
+; CHECK-NEXT: lsr w12, w12, #24
+; CHECK-NEXT: umov w13, v1.b[12]
+; CHECK-NEXT: extr w9, w11, w9, #2
+; CHECK-NEXT: cmp w11, #3
+; CHECK-NEXT: umov w11, v0.b[13]
+; CHECK-NEXT: lsl w10, w10, #24
+; CHECK-NEXT: mov v2.b[9], w12
+; CHECK-NEXT: lsr x12, x8, #32
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: lsr w9, w9, #24
+; CHECK-NEXT: umull x10, w10, w13
+; CHECK-NEXT: umov w13, v1.b[13]
+; CHECK-NEXT: extr w8, w12, w8, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: lsl w11, w11, #24
+; CHECK-NEXT: mov v2.b[10], w9
+; CHECK-NEXT: umov w9, v0.b[14]
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr x12, x10, #32
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: umull x11, w11, w13
+; CHECK-NEXT: umov w13, v1.b[14]
+; CHECK-NEXT: extr w10, w12, w10, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: mov v2.b[11], w8
+; CHECK-NEXT: lsl w8, w9, #24
+; CHECK-NEXT: umov w9, v0.b[15]
+; CHECK-NEXT: csinv w10, w10, wzr, ls
+; CHECK-NEXT: lsr x12, x11, #32
+; CHECK-NEXT: lsr w10, w10, #24
+; CHECK-NEXT: umull x8, w8, w13
+; CHECK-NEXT: umov w13, v1.b[15]
+; CHECK-NEXT: extr w11, w12, w11, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: mov v2.b[12], w10
+; CHECK-NEXT: lsl w9, w9, #24
+; CHECK-NEXT: csinv w10, w11, wzr, ls
+; CHECK-NEXT: lsr x11, x8, #32
+; CHECK-NEXT: umull x9, w9, w13
+; CHECK-NEXT: lsr w10, w10, #24
+; CHECK-NEXT: extr w8, w11, w8, #2
+; CHECK-NEXT: cmp w11, #3
+; CHECK-NEXT: mov v2.b[13], w10
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr x10, x9, #32
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: extr w9, w10, w9, #2
+; CHECK-NEXT: cmp w10, #3
+; CHECK-NEXT: mov v2.b[14], w8
+; CHECK-NEXT: csinv w8, w9, wzr, ls
+; CHECK-NEXT: lsr w8, w8, #24
+; CHECK-NEXT: mov v2.b[15], w8
+; CHECK-NEXT: mov v0.16b, v2.16b
+; CHECK-NEXT: ret
+ %tmp = call <16 x i8> @llvm.umul.fix.sat.v16i8(<16 x i8> %x, <16 x i8> %y, i32 2)
+ ret <16 x i8> %tmp
+}
+
+define <4 x i16> @vec_v4i16(<4 x i16> %x, <4 x i16> %y) {
+; CHECK-LABEL: vec_v4i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: umov w8, v0.h[0]
+; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NEXT: umov w10, v1.h[0]
+; CHECK-NEXT: umov w9, v0.h[1]
+; CHECK-NEXT: umov w11, v1.h[1]
+; CHECK-NEXT: umov w12, v0.h[2]
+; CHECK-NEXT: umov w13, v0.h[3]
+; CHECK-NEXT: lsl w8, w8, #16
+; CHECK-NEXT: lsl w9, w9, #16
+; CHECK-NEXT: lsl w12, w12, #16
+; CHECK-NEXT: umull x8, w8, w10
+; CHECK-NEXT: umov w10, v1.h[2]
+; CHECK-NEXT: umull x9, w9, w11
+; CHECK-NEXT: lsr x11, x8, #32
+; CHECK-NEXT: lsr x14, x9, #32
+; CHECK-NEXT: umull x10, w12, w10
+; CHECK-NEXT: umov w12, v1.h[3]
+; CHECK-NEXT: extr w8, w11, w8, #10
+; CHECK-NEXT: cmp w11, #1023
+; CHECK-NEXT: lsl w11, w13, #16
+; CHECK-NEXT: extr w9, w14, w9, #10
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: cmp w14, #1023
+; CHECK-NEXT: lsr x13, x10, #32
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: umull x11, w11, w12
+; CHECK-NEXT: lsr w9, w9, #16
+; CHECK-NEXT: cmp w13, #1023
+; CHECK-NEXT: fmov s0, w8
+; CHECK-NEXT: extr w8, w13, w10, #10
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: mov v0.h[1], w9
+; CHECK-NEXT: lsr x9, x11, #32
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: extr w10, w9, w11, #10
+; CHECK-NEXT: cmp w9, #1023
+; CHECK-NEXT: mov v0.h[2], w8
+; CHECK-NEXT: csinv w8, w10, wzr, ls
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: mov v0.h[3], w8
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT: ret
+ %tmp = call <4 x i16> @llvm.umul.fix.sat.v4i16(<4 x i16> %x, <4 x i16> %y, i32 10)
+ ret <4 x i16> %tmp
+}
+
+define <8 x i16> @vec_v8i16(<8 x i16> %x, <8 x i16> %y) {
+; CHECK-LABEL: vec_v8i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: umov w8, v0.h[1]
+; CHECK-NEXT: umov w9, v0.h[0]
+; CHECK-NEXT: umov w10, v1.h[1]
+; CHECK-NEXT: umov w11, v1.h[0]
+; CHECK-NEXT: umov w12, v0.h[2]
+; CHECK-NEXT: umov w13, v0.h[3]
+; CHECK-NEXT: lsl w8, w8, #16
+; CHECK-NEXT: lsl w9, w9, #16
+; CHECK-NEXT: umull x8, w8, w10
+; CHECK-NEXT: umov w10, v1.h[2]
+; CHECK-NEXT: umull x9, w9, w11
+; CHECK-NEXT: lsl w11, w12, #16
+; CHECK-NEXT: lsr x12, x8, #32
+; CHECK-NEXT: lsr x14, x9, #32
+; CHECK-NEXT: umull x10, w11, w10
+; CHECK-NEXT: umov w11, v1.h[3]
+; CHECK-NEXT: extr w8, w12, w8, #2
+; CHECK-NEXT: cmp w12, #3
+; CHECK-NEXT: lsl w12, w13, #16
+; CHECK-NEXT: extr w9, w14, w9, #2
+; CHECK-NEXT: umov w13, v0.h[4]
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: lsr x14, x10, #32
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: umull x11, w12, w11
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: lsr w9, w9, #16
+; CHECK-NEXT: extr w10, w14, w10, #2
+; CHECK-NEXT: cmp w14, #3
+; CHECK-NEXT: umov w12, v0.h[5]
+; CHECK-NEXT: fmov s2, w9
+; CHECK-NEXT: umov w9, v1.h[4]
+; CHECK-NEXT: csinv w10, w10, wzr, ls
+; CHECK-NEXT: lsr w10, w10, #16
+; CHECK-NEXT: mov v2.h[1], w8
+; CHECK-NEXT: lsl w8, w13, #16
+; CHECK-NEXT: lsr x13, x11, #32
+; CHECK-NEXT: lsl w12, w12, #16
+; CHECK-NEXT: umull x8, w8, w9
+; CHECK-NEXT: umov w9, v1.h[5]
+; CHECK-NEXT: extr w11, w13, w11, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.h[2], w10
+; CHECK-NEXT: umov w10, v0.h[6]
+; CHECK-NEXT: csinv w11, w11, wzr, ls
+; CHECK-NEXT: lsr x13, x8, #32
+; CHECK-NEXT: lsr w11, w11, #16
+; CHECK-NEXT: umull x9, w12, w9
+; CHECK-NEXT: umov w12, v1.h[6]
+; CHECK-NEXT: extr w8, w13, w8, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.h[3], w11
+; CHECK-NEXT: lsl w10, w10, #16
+; CHECK-NEXT: umov w11, v0.h[7]
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr x13, x9, #32
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: umull x10, w10, w12
+; CHECK-NEXT: umov w12, v1.h[7]
+; CHECK-NEXT: extr w9, w13, w9, #2
+; CHECK-NEXT: cmp w13, #3
+; CHECK-NEXT: mov v2.h[4], w8
+; CHECK-NEXT: lsl w8, w11, #16
+; CHECK-NEXT: csinv w9, w9, wzr, ls
+; CHECK-NEXT: lsr x11, x10, #32
+; CHECK-NEXT: umull x8, w8, w12
+; CHECK-NEXT: lsr w9, w9, #16
+; CHECK-NEXT: extr w10, w11, w10, #2
+; CHECK-NEXT: cmp w11, #3
+; CHECK-NEXT: mov v2.h[5], w9
+; CHECK-NEXT: csinv w9, w10, wzr, ls
+; CHECK-NEXT: lsr x10, x8, #32
+; CHECK-NEXT: lsr w9, w9, #16
+; CHECK-NEXT: extr w8, w10, w8, #2
+; CHECK-NEXT: cmp w10, #3
+; CHECK-NEXT: mov v2.h[6], w9
+; CHECK-NEXT: csinv w8, w8, wzr, ls
+; CHECK-NEXT: lsr w8, w8, #16
+; CHECK-NEXT: mov v2.h[7], w8
+; CHECK-NEXT: mov v0.16b, v2.16b
+; CHECK-NEXT: ret
+ %tmp = call <8 x i16> @llvm.umul.fix.sat.v8i16(<8 x i16> %x, <8 x i16> %y, i32 2)
+ ret <8 x i16> %tmp
+}
+
+define <2 x i32> @vec_v2i32(<2 x i32> %x, <2 x i32> %y) {
+; CHECK-LABEL: vec_v2i32:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
@@ -104,40 +551,128 @@ define <2 x i32> @vec(<2 x i32> %x, <2 x i32> %y) nounwind {
ret <2 x i32> %tmp
}
-define <4 x i32> @vec2(<4 x i32> %x, <4 x i32> %y) nounwind {
-; CHECK-LABEL: vec2:
+define <4 x i32> @vec_v4i32(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: vec_v4i32:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, v1.s[1]
; CHECK-NEXT: mov w9, v0.s[1]
; CHECK-NEXT: fmov w10, s0
; CHECK-NEXT: mov w11, v0.s[2]
-; CHECK-NEXT: mov w13, v0.s[3]
-; CHECK-NEXT: mov w12, v1.s[3]
+; CHECK-NEXT: mov w14, v1.s[3]
+; CHECK-NEXT: mov w15, v0.s[3]
; CHECK-NEXT: umull x8, w9, w8
; CHECK-NEXT: fmov w9, s1
; CHECK-NEXT: umull x9, w10, w9
-; CHECK-NEXT: tst x8, #0xffffffff00000000
+; CHECK-NEXT: lsr x12, x8, #32
; CHECK-NEXT: mov w10, v1.s[2]
+; CHECK-NEXT: extr w8, w12, w8, #15
+; CHECK-NEXT: cmp w12, #8, lsl #12 // =32768
+; CHECK-NEXT: lsr x13, x9, #32
+; CHECK-NEXT: csinv w8, w8, wzr, lo
+; CHECK-NEXT: umull x10, w11, w10
+; CHECK-NEXT: extr w9, w13, w9, #15
+; CHECK-NEXT: cmp w13, #8, lsl #12 // =32768
+; CHECK-NEXT: csinv w9, w9, wzr, lo
+; CHECK-NEXT: fmov s0, w9
+; CHECK-NEXT: lsr x9, x10, #32
+; CHECK-NEXT: extr w10, w9, w10, #15
+; CHECK-NEXT: cmp w9, #8, lsl #12 // =32768
+; CHECK-NEXT: mov v0.s[1], w8
+; CHECK-NEXT: umull x8, w15, w14
+; CHECK-NEXT: csinv w9, w10, wzr, lo
+; CHECK-NEXT: lsr x10, x8, #32
+; CHECK-NEXT: mov v0.s[2], w9
+; CHECK-NEXT: extr w8, w10, w8, #15
+; CHECK-NEXT: cmp w10, #8, lsl #12 // =32768
+; CHECK-NEXT: csinv w8, w8, wzr, lo
+; CHECK-NEXT: mov v0.s[3], w8
+; CHECK-NEXT: ret
+ %tmp = call <4 x i32> @llvm.umul.fix.sat.v4i32(<4 x i32> %x, <4 x i32> %y, i32 15)
+ ret <4 x i32> %tmp
+}
+
+define <8 x i32> @vec_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; CHECK-LABEL: vec_v8i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov w8, v2.s[1]
+; CHECK-NEXT: mov w9, v0.s[1]
+; CHECK-NEXT: mov w10, v2.s[2]
+; CHECK-NEXT: mov w11, v0.s[2]
+; CHECK-NEXT: mov w12, v2.s[3]
+; CHECK-NEXT: fmov w13, s2
+; CHECK-NEXT: mov w14, v0.s[3]
+; CHECK-NEXT: fmov w15, s0
+; CHECK-NEXT: mov w16, v3.s[1]
+; CHECK-NEXT: mov w17, v1.s[1]
+; CHECK-NEXT: umull x8, w9, w8
+; CHECK-NEXT: umull x9, w15, w13
+; CHECK-NEXT: fmov w15, s1
+; CHECK-NEXT: mov w13, v3.s[2]
+; CHECK-NEXT: umull x10, w11, w10
+; CHECK-NEXT: umull x11, w14, w12
+; CHECK-NEXT: fmov w14, s3
+; CHECK-NEXT: tst x8, #0xffffffff00000000
+; CHECK-NEXT: umull x12, w17, w16
; CHECK-NEXT: csinv w8, w8, wzr, eq
; CHECK-NEXT: tst x9, #0xffffffff00000000
; CHECK-NEXT: csinv w9, w9, wzr, eq
+; CHECK-NEXT: tst x10, #0xffffffff00000000
+; CHECK-NEXT: mov w17, v1.s[3]
+; CHECK-NEXT: umull x14, w15, w14
+; CHECK-NEXT: csinv w10, w10, wzr, eq
+; CHECK-NEXT: tst x11, #0xffffffff00000000
+; CHECK-NEXT: csinv w11, w11, wzr, eq
+; CHECK-NEXT: tst x12, #0xffffffff00000000
+; CHECK-NEXT: mov w15, v1.s[2]
+; CHECK-NEXT: csinv w12, w12, wzr, eq
; CHECK-NEXT: fmov s0, w9
-; CHECK-NEXT: umull x9, w11, w10
+; CHECK-NEXT: mov w16, v3.s[3]
+; CHECK-NEXT: tst x14, #0xffffffff00000000
+; CHECK-NEXT: csinv w14, w14, wzr, eq
+; CHECK-NEXT: fmov s1, w14
+; CHECK-NEXT: umull x9, w15, w13
; CHECK-NEXT: mov v0.s[1], w8
+; CHECK-NEXT: umull x8, w17, w16
+; CHECK-NEXT: mov v1.s[1], w12
; CHECK-NEXT: tst x9, #0xffffffff00000000
-; CHECK-NEXT: csinv w8, w9, wzr, eq
-; CHECK-NEXT: umull x9, w13, w12
-; CHECK-NEXT: mov v0.s[2], w8
-; CHECK-NEXT: tst x9, #0xffffffff00000000
-; CHECK-NEXT: csinv w8, w9, wzr, eq
-; CHECK-NEXT: mov v0.s[3], w8
+; CHECK-NEXT: csinv w9, w9, wzr, eq
+; CHECK-NEXT: mov v0.s[2], w10
+; CHECK-NEXT: tst x8, #0xffffffff00000000
+; CHECK-NEXT: csinv w8, w8, wzr, eq
+; CHECK-NEXT: mov v1.s[2], w9
+; CHECK-NEXT: mov v0.s[3], w11
+; CHECK-NEXT: mov v1.s[3], w8
; CHECK-NEXT: ret
- %tmp = call <4 x i32> @llvm.umul.fix.sat.v4i32(<4 x i32> %x, <4 x i32> %y, i32 0)
- ret <4 x i32> %tmp
+ %tmp = call <8 x i32> @llvm.umul.fix.sat.v8i32(<8 x i32> %x, <8 x i32> %y, i32 0)
+ ret <8 x i32> %tmp
}
-define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
-; CHECK-LABEL: vec3:
+define <2 x i64> @vec_v2i64(<2 x i64> %x, <2 x i64> %y) {
+; CHECK-LABEL: vec_v2i64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov x8, v1.d[1]
+; CHECK-NEXT: mov x9, v0.d[1]
+; CHECK-NEXT: fmov x11, d0
+; CHECK-NEXT: mul x10, x9, x8
+; CHECK-NEXT: umulh x8, x9, x8
+; CHECK-NEXT: fmov x9, d1
+; CHECK-NEXT: mul x12, x11, x9
+; CHECK-NEXT: umulh x9, x11, x9
+; CHECK-NEXT: extr x10, x8, x10, #15
+; CHECK-NEXT: cmp x8, #8, lsl #12 // =32768
+; CHECK-NEXT: csinv x10, x10, xzr, lo
+; CHECK-NEXT: extr x8, x9, x12, #15
+; CHECK-NEXT: cmp x9, #8, lsl #12 // =32768
+; CHECK-NEXT: csinv x8, x8, xzr, lo
+; CHECK-NEXT: fmov d0, x8
+; CHECK-NEXT: mov v0.d[1], x10
+; CHECK-NEXT: ret
+ %tmp = call <2 x i64> @llvm.umul.fix.sat.v2i64(<2 x i64> %x, <2 x i64> %y, i32 15)
+ ret <2 x i64> %tmp
+}
+
+define <4 x i64> @vec_v4i64(<4 x i64> %x, <4 x i64> %y) {
+; CHECK-LABEL: vec_v4i64:
; CHECK: // %bb.0:
; CHECK-NEXT: mov x8, v2.d[1]
; CHECK-NEXT: mov x9, v0.d[1]
@@ -177,30 +712,6 @@ define <4 x i64> @vec3(<4 x i64> %x, <4 x i64> %y) nounwind {
ret <4 x i64> %tmp
}
-define i64 @func7(i64 %x, i64 %y) nounwind {
-; CHECK-LABEL: func7:
-; CHECK: // %bb.0:
-; CHECK-NEXT: mul x9, x0, x1
-; CHECK-NEXT: mov w8, #-1 // =0xffffffff
-; CHECK-NEXT: umulh x10, x0, x1
-; CHECK-NEXT: extr x9, x10, x9, #32
-; CHECK-NEXT: cmp x10, x8
-; CHECK-NEXT: csinv x0, x9, xzr, ls
-; CHECK-NEXT: ret
- %tmp = call i64 @llvm.umul.fix.sat.i64(i64 %x, i64 %y, i32 32)
- ret i64 %tmp
-}
-
-define i64 @func8(i64 %x, i64 %y) nounwind {
-; CHECK-LABEL: func8:
-; CHECK: // %bb.0:
-; CHECK-NEXT: mul x9, x0, x1
-; CHECK-NEXT: mov x8, #9223372036854775807 // =0x7fffffffffffffff
-; CHECK-NEXT: umulh x10, x0, x1
-; CHECK-NEXT: extr x9, x10, x9, #63
-; CHECK-NEXT: cmp x10, x8
-; CHECK-NEXT: csinv x0, x9, xzr, ls
-; CHECK-NEXT: ret
- %tmp = call i64 @llvm.umul.fix.sat.i64(i64 %x, i64 %y, i32 63)
- ret i64 %tmp
-}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK-GI: {{.*}}
+; CHECK-SD: {{.*}}
More information about the llvm-commits
mailing list