mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-12-05 02:07:56 +00:00
bc25529f39
This teaches the IRTranslator to emit G_BSWAP when it runs into Intrinsic::bswap. This allows us to select G_BSWAP for non-vector types in AArch64. Add a select-bswap.mir test, and add global isel checks to a couple existing tests in test/CodeGen/AArch64. This doesn't handle every bswap case, since some of these rely on known bits stuff. This just lets us handle the naive case. Differential Revision: https://reviews.llvm.org/D58081 llvm-svn: 353861
159 lines
4.8 KiB
LLVM
159 lines
4.8 KiB
LLVM
; RUN: llc -verify-machineinstrs -o - %s -mtriple=aarch64-linux-gnu | FileCheck %s
|
|
; RUN: llc -global-isel -global-isel-abort=2 -pass-remarks-missed=gisel* -verify-machineinstrs -o - %s -mtriple=aarch64-linux-gnu | FileCheck %s --check-prefixes=FALLBACK,GISEL
|
|
|
|
@var32 = global i32 0
|
|
@var64 = global i64 0
|
|
|
|
; FALLBACK-NOT: remark{{.*}}rev_i32
|
|
define void @rev_i32() {
|
|
; CHECK-LABEL: rev_i32:
|
|
; GISEL-LABEL: rev_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val1_tmp = call i32 @llvm.bswap.i32(i32 %val0_tmp)
|
|
; CHECK: rev {{w[0-9]+}}, {{w[0-9]+}}
|
|
; GISEL: rev {{w[0-9]+}}, {{w[0-9]+}}
|
|
store volatile i32 %val1_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
; FALLBACK-NOT: remark{{.*}}rev_i64
|
|
define void @rev_i64() {
|
|
; CHECK-LABEL: rev_i64:
|
|
; GISEL-LABEL: rev_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val1_tmp = call i64 @llvm.bswap.i64(i64 %val0_tmp)
|
|
; CHECK: rev {{x[0-9]+}}, {{x[0-9]+}}
|
|
; GISEL: rev {{x[0-9]+}}, {{x[0-9]+}}
|
|
store volatile i64 %val1_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
define void @rev32_i64() {
|
|
; CHECK-LABEL: rev32_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val1_tmp = shl i64 %val0_tmp, 32
|
|
%val5_tmp = sub i64 64, 32
|
|
%val2_tmp = lshr i64 %val0_tmp, %val5_tmp
|
|
%val3_tmp = or i64 %val1_tmp, %val2_tmp
|
|
%val4_tmp = call i64 @llvm.bswap.i64(i64 %val3_tmp)
|
|
; CHECK: rev32 {{x[0-9]+}}, {{x[0-9]+}}
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
define void @rev16_i32() {
|
|
; CHECK-LABEL: rev16_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val1_tmp = shl i32 %val0_tmp, 16
|
|
%val2_tmp = lshr i32 %val0_tmp, 16
|
|
%val3_tmp = or i32 %val1_tmp, %val2_tmp
|
|
%val4_tmp = call i32 @llvm.bswap.i32(i32 %val3_tmp)
|
|
; CHECK: rev16 {{w[0-9]+}}, {{w[0-9]+}}
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @clz_zerodef_i32() {
|
|
; CHECK-LABEL: clz_zerodef_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val4_tmp = call i32 @llvm.ctlz.i32(i32 %val0_tmp, i1 0)
|
|
; CHECK: clz {{w[0-9]+}}, {{w[0-9]+}}
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @clz_zerodef_i64() {
|
|
; CHECK-LABEL: clz_zerodef_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val4_tmp = call i64 @llvm.ctlz.i64(i64 %val0_tmp, i1 0)
|
|
; CHECK: clz {{x[0-9]+}}, {{x[0-9]+}}
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
define void @clz_zeroundef_i32() {
|
|
; CHECK-LABEL: clz_zeroundef_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val4_tmp = call i32 @llvm.ctlz.i32(i32 %val0_tmp, i1 1)
|
|
; CHECK: clz {{w[0-9]+}}, {{w[0-9]+}}
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @clz_zeroundef_i64() {
|
|
; CHECK-LABEL: clz_zeroundef_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val4_tmp = call i64 @llvm.ctlz.i64(i64 %val0_tmp, i1 1)
|
|
; CHECK: clz {{x[0-9]+}}, {{x[0-9]+}}
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
define void @cttz_zerodef_i32() {
|
|
; CHECK-LABEL: cttz_zerodef_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val4_tmp = call i32 @llvm.cttz.i32(i32 %val0_tmp, i1 0)
|
|
; CHECK: rbit [[REVERSED:w[0-9]+]], {{w[0-9]+}}
|
|
; CHECK: clz {{w[0-9]+}}, [[REVERSED]]
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @cttz_zerodef_i64() {
|
|
; CHECK-LABEL: cttz_zerodef_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val4_tmp = call i64 @llvm.cttz.i64(i64 %val0_tmp, i1 0)
|
|
; CHECK: rbit [[REVERSED:x[0-9]+]], {{x[0-9]+}}
|
|
; CHECK: clz {{x[0-9]+}}, [[REVERSED]]
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
define void @cttz_zeroundef_i32() {
|
|
; CHECK-LABEL: cttz_zeroundef_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val4_tmp = call i32 @llvm.cttz.i32(i32 %val0_tmp, i1 1)
|
|
; CHECK: rbit [[REVERSED:w[0-9]+]], {{w[0-9]+}}
|
|
; CHECK: clz {{w[0-9]+}}, [[REVERSED]]
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @cttz_zeroundef_i64() {
|
|
; CHECK-LABEL: cttz_zeroundef_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val4_tmp = call i64 @llvm.cttz.i64(i64 %val0_tmp, i1 1)
|
|
; CHECK: rbit [[REVERSED:x[0-9]+]], {{x[0-9]+}}
|
|
; CHECK: clz {{x[0-9]+}}, [[REVERSED]]
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
; These two are just compilation tests really: the operation's set to Expand in
|
|
; ISelLowering.
|
|
define void @ctpop_i32() {
|
|
; CHECK-LABEL: ctpop_i32:
|
|
%val0_tmp = load i32, i32* @var32
|
|
%val4_tmp = call i32 @llvm.ctpop.i32(i32 %val0_tmp)
|
|
store volatile i32 %val4_tmp, i32* @var32
|
|
ret void
|
|
}
|
|
|
|
define void @ctpop_i64() {
|
|
; CHECK-LABEL: ctpop_i64:
|
|
%val0_tmp = load i64, i64* @var64
|
|
%val4_tmp = call i64 @llvm.ctpop.i64(i64 %val0_tmp)
|
|
store volatile i64 %val4_tmp, i64* @var64
|
|
ret void
|
|
}
|
|
|
|
|
|
declare i32 @llvm.bswap.i32(i32)
|
|
declare i64 @llvm.bswap.i64(i64)
|
|
declare i32 @llvm.ctlz.i32 (i32, i1)
|
|
declare i64 @llvm.ctlz.i64 (i64, i1)
|
|
declare i32 @llvm.cttz.i32 (i32, i1)
|
|
declare i64 @llvm.cttz.i64 (i64, i1)
|
|
declare i32 @llvm.ctpop.i32 (i32)
|
|
declare i64 @llvm.ctpop.i64 (i64)
|