useAA significantly improves the handling of vector code that has TBAA
information attached. It also helps other cases, as shown by the testsuite
changes here. The only real downside I've seen is that it interferes with
MergeConsecutiveStores. The problem is that that optimization works top
down, starting at the first store in the chain, and looks for cases where
the chain result is only used by a single related store. These related
stores don't alias, so useAA will have rewritten all the later stores to
use a different chain input (typically the same one as the first store).
I think the advantages outweigh the disadvantages though, so for now I've
just disabled alias analysis for the unaligned-01.ll test.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193521
91177308-0d34-0410-b5e6-
96231b3b80d8
SystemZSubtarget(const std::string &TT, const std::string &CPU,
const std::string &FS);
+ // This is important for reducing register pressure in vector code.
+ virtual bool useAA() const LLVM_OVERRIDE { return true; }
+
// Automatically generated by tblgen.
void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
; Test 32-bit ANDs in which the second operand is variable.
;
-; RUN: llc < %s -mtriple=s390x-linux-gnu -combiner-alias-analysis \
-; RUN: -combiner-global-alias-analysis | FileCheck %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
; Check that there are no spills.
define void @f1(<16 x i32> *%src1, <16 x float> *%dest) {
; Test 32-bit byteswaps from memory to registers.
;
-; RUN: llc < %s -mtriple=s390x-linux-gnu -combiner-alias-analysis \
-; RUN: -combiner-global-alias-analysis | FileCheck %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
declare i32 @llvm.bswap.i32(i32 %a)
define void @f1(i128 *%aptr) {
; CHECK-LABEL: f1:
; CHECK: algfi {{%r[0-5]}}, 1
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 128
define void @f2(i128 *%aptr) {
; CHECK-LABEL: f2:
; CHECK: algfi {{%r[0-5]}}, 4294967295
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 128
define void @f3(i128 *%aptr) {
; CHECK-LABEL: f3:
; CHECK: algr
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 128
define void @f4(i128 *%aptr) {
; CHECK-LABEL: f4:
; CHECK: algr
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 128
define void @f1(i128 *%aptr, i32 %b) {
; CHECK-LABEL: f1:
; CHECK: algfr {{%r[0-5]}}, %r3
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f2(i128 *%aptr, i64 %b) {
; CHECK-LABEL: f2:
; CHECK: algfr {{%r[0-5]}}, %r3
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f3(i128 *%aptr, i64 %b) {
; CHECK-LABEL: f3:
; CHECK: algfr {{%r[0-5]}}, %r3
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f4(i128 *%aptr, i32 *%bsrc) {
; CHECK-LABEL: f4:
; CHECK: algf {{%r[0-5]}}, 0(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f5(i128 *%aptr, i32 *%bsrc) {
; CHECK-LABEL: f5:
; CHECK: algf {{%r[0-5]}}, 524284(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
; CHECK-LABEL: f6:
; CHECK: agfi %r3, 524288
; CHECK: algf {{%r[0-5]}}, 0(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f7(i128 *%aptr, i32 *%bsrc) {
; CHECK-LABEL: f7:
; CHECK: algf {{%r[0-5]}}, -4(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
define void @f8(i128 *%aptr, i32 *%bsrc) {
; CHECK-LABEL: f8:
; CHECK: algf {{%r[0-5]}}, -524288(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
; CHECK-LABEL: f9:
; CHECK: agfi %r3, -524292
; CHECK: algf {{%r[0-5]}}, 0(%r3)
-; CHECK: alcgr
+; CHECK: alcg
; CHECK: br %r14
%a = load i128 *%aptr
%xor = xor i128 %a, 127
; Check that unaligned accesses are allowed in general. We check the
; few exceptions (like CRL) in their respective test files.
;
-; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
+; FIXME: -combiner-alias-analysis (the default for SystemZ) stops
+; f1 from being optimized.
+; RUN: llc < %s -mtriple=s390x-linux-gnu -combiner-alias-analysis=false \
+; RUN: | FileCheck %s
; Check that these four byte stores become a single word store.
define void @f1(i8 *%ptr) {