Some of the tests from the X86 directory can be generalized to improve coverage for other architectures (cont.)
51 lines
2.4 KiB
LLVM
51 lines
2.4 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: %if x86-registered-target %{ opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux < %s -slp-threshold=-10 | FileCheck %s %}
|
|
; RUN: %if aarch64-registered-target %{ opt -S --passes=slp-vectorizer -mtriple=aarch64-unknown-linux < %s -slp-threshold=-10 | FileCheck %s %}
|
|
|
|
define i64 @src(i32 %a) {
|
|
; CHECK-LABEL: define i64 @src(
|
|
; CHECK-SAME: i32 [[A:%.*]]) {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: [[TMP17:%.*]] = sext i32 [[A]] to i64
|
|
; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> poison, i32 [[A]], i32 0
|
|
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer
|
|
; CHECK-NEXT: [[TMP3:%.*]] = sext <4 x i32> [[TMP2]] to <4 x i64>
|
|
; CHECK-NEXT: [[TMP4:%.*]] = add nsw <4 x i64> [[TMP3]], <i64 4294967297, i64 4294967297, i64 4294967297, i64 4294967297>
|
|
; CHECK-NEXT: [[TMP6:%.*]] = and <4 x i64> [[TMP4]], <i64 1, i64 1, i64 1, i64 1>
|
|
; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[TMP6]])
|
|
; CHECK-NEXT: [[TMP16:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[TMP4]])
|
|
; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> poison, i64 [[TMP16]], i32 0
|
|
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i64> [[TMP8]], i64 [[TMP18]], i32 1
|
|
; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x i64> <i64 poison, i64 4294967297>, i64 [[TMP17]], i32 0
|
|
; CHECK-NEXT: [[TMP11:%.*]] = add <2 x i64> [[TMP9]], [[TMP10]]
|
|
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i64> [[TMP11]], i32 0
|
|
; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP11]], i32 1
|
|
; CHECK-NEXT: [[TMP21:%.*]] = add i64 [[TMP12]], [[TMP13]]
|
|
; CHECK-NEXT: ret i64 [[TMP21]]
|
|
;
|
|
entry:
|
|
%0 = sext i32 %a to i64
|
|
%1 = add nsw i64 %0, 4294967297
|
|
%2 = sext i32 %a to i64
|
|
%3 = add nsw i64 %2, 4294967297
|
|
%4 = add i64 %3, %1
|
|
%5 = and i64 %3, 1
|
|
%6 = add i64 %4, %5
|
|
%7 = sext i32 %a to i64
|
|
%8 = add nsw i64 %7, 4294967297
|
|
%9 = add i64 %8, %6
|
|
%10 = and i64 %8, 1
|
|
%11 = add i64 %9, %10
|
|
%12 = sext i32 %a to i64
|
|
%13 = add nsw i64 %12, 4294967297
|
|
%14 = add i64 %13, %11
|
|
%15 = and i64 %13, 1
|
|
%16 = add i64 %14, %15
|
|
%17 = sext i32 %a to i64
|
|
%18 = add nsw i64 %17, 4294967297
|
|
%19 = add i64 %18, %16
|
|
%20 = and i64 %18, 1
|
|
%21 = add i64 %19, %20
|
|
ret i64 %21
|
|
}
|