mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-01 08:23:21 +01:00
77f91408dd
Unlike most -- hopefully "all other", but I'm still checking -- memory
instructions we support, LOAD REVERSED and STORE REVERSED may access
the memory location several times. This means that they are not suitable
for volatile loads and stores.
This patch is a prerequisite for better atomic load and store support.
The same principle applies there: almost all memory instructions we
support are inherently atomic ("block concurrent"), but LOAD REVERSED
and STORE REVERSED are exceptions.
Other instructions continue to allow volatile operands. I will add
positive "allows volatile" tests at the same time as the "allows atomic
load or store" tests.
llvm-svn: 183002
100 lines
2.5 KiB
LLVM
100 lines
2.5 KiB
LLVM
; Test 64-bit byteswaps from memory to registers.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
|
|
|
declare i64 @llvm.bswap.i64(i64 %a)
|
|
|
|
; Check LRVG with no displacement.
|
|
define i64 @f1(i64 *%src) {
|
|
; CHECK: f1:
|
|
; CHECK: lrvg %r2, 0(%r2)
|
|
; CHECK: br %r14
|
|
%a = load i64 *%src
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check the high end of the aligned LRVG range.
|
|
define i64 @f2(i64 *%src) {
|
|
; CHECK: f2:
|
|
; CHECK: lrvg %r2, 524280(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i64 *%src, i64 65535
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check the next doubleword up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define i64 @f3(i64 *%src) {
|
|
; CHECK: f3:
|
|
; CHECK: agfi %r2, 524288
|
|
; CHECK: lrvg %r2, 0(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i64 *%src, i64 65536
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check the high end of the negative aligned LRVG range.
|
|
define i64 @f4(i64 *%src) {
|
|
; CHECK: f4:
|
|
; CHECK: lrvg %r2, -8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i64 *%src, i64 -1
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check the low end of the LRVG range.
|
|
define i64 @f5(i64 *%src) {
|
|
; CHECK: f5:
|
|
; CHECK: lrvg %r2, -524288(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i64 *%src, i64 -65536
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check the next doubleword down, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define i64 @f6(i64 *%src) {
|
|
; CHECK: f6:
|
|
; CHECK: agfi %r2, -524296
|
|
; CHECK: lrvg %r2, 0(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i64 *%src, i64 -65537
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check that LRVG allows an index.
|
|
define i64 @f7(i64 %src, i64 %index) {
|
|
; CHECK: f7:
|
|
; CHECK: lrvg %r2, 524287({{%r3,%r2|%r2,%r3}})
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 524287
|
|
%ptr = inttoptr i64 %add2 to i64 *
|
|
%a = load i64 *%ptr
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|
|
|
|
; Check that volatile accesses do not use LRVG, which might access the
|
|
; storage multple times.
|
|
define i64 @f8(i64 *%src) {
|
|
; CHECK: f8:
|
|
; CHECK: lg [[REG:%r[0-5]]], 0(%r2)
|
|
; CHECK: lrvgr %r2, [[REG]]
|
|
; CHECK: br %r14
|
|
%a = load volatile i64 *%src
|
|
%swapped = call i64 @llvm.bswap.i64(i64 %a)
|
|
ret i64 %swapped
|
|
}
|