mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-24 13:06:56 +00:00
24517d023f
The current memory-instruction optimization logic in CGP, which sinks parts of
the address computation that can be adsorbed by the addressing mode, does this
by explicitly converting the relevant part of the address computation into
IR-level integer operations (making use of ptrtoint and inttoptr). For most
targets this is currently not a problem, but for targets wishing to make use of
IR-level aliasing analysis during CodeGen, the use of ptrtoint/inttoptr is a
problem for two reasons:
1. BasicAA becomes less powerful in the face of the ptrtoint/inttoptr
2. In cases where type-punning was used, and BasicAA was used
to override TBAA, BasicAA may no longer do so. (this had forced us to disable
all use of TBAA in CodeGen; something which we can now enable again)
This (use of GEPs instead of ptrtoint/inttoptr) is not currently enabled by
default (except for those targets that use AA during CodeGen), and so aside
from some PowerPC subtargets and SystemZ, there should be no change in
behavior. We may be able to switch completely away from the ptrtoint/inttoptr
sinking on all targets, but further testing is required.
I've doubled-up on a number of existing tests that are sensitive to the
address sinking behavior (including some store-merging tests that are
sensitive to the order of the resulting ADD operations at the SDAG level).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206092 91177308-0d34-0410-b5e6-96231b3b80d8
324 lines
13 KiB
LLVM
324 lines
13 KiB
LLVM
; RUN: opt -S -codegenprepare %s -o - | FileCheck %s
|
|
; RUN: opt -S -codegenprepare -addr-sink-using-gep=1 %s -o - | FileCheck -check-prefix=CHECK-GEP %s
|
|
; This file tests the different cases what are involved when codegen prepare
|
|
; tries to get sign extension out of the way of addressing mode.
|
|
; This tests require an actual target as addressing mode decisions depends
|
|
; on the target.
|
|
|
|
target datalayout = "e-i64:64-f80:128-s:64-n8:16:32:64-S128"
|
|
target triple = "x86_64-apple-macosx"
|
|
|
|
|
|
; Check that we correctly promote both operands of the promotable add.
|
|
; CHECK-LABEL: @twoArgsPromotion
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i32 %arg1 to i64
|
|
; CHECK: [[ARG2SEXT:%[a-zA-Z_0-9-]+]] = sext i32 %arg2 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], [[ARG2SEXT]]
|
|
; CHECK: inttoptr i64 [[PROMOTED]] to i8*
|
|
; CHECK: ret
|
|
define i8 @twoArgsPromotion(i32 %arg1, i32 %arg2) {
|
|
%add = add nsw i32 %arg1, %arg2
|
|
%sextadd = sext i32 %add to i64
|
|
%base = inttoptr i64 %sextadd to i8*
|
|
%res = load i8* %base
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we do not promote both operands of the promotable add when
|
|
; the instruction will not be folded into the addressing mode.
|
|
; Otherwise, we will increase the number of instruction executed.
|
|
; (This is a heuristic of course, because the new sext could have been
|
|
; merged with something else.)
|
|
; CHECK-LABEL: @twoArgsNoPromotion
|
|
; CHECK: add nsw i32 %arg1, %arg2
|
|
; CHECK: ret
|
|
define i8 @twoArgsNoPromotion(i32 %arg1, i32 %arg2, i8* %base) {
|
|
%add = add nsw i32 %arg1, %arg2
|
|
%sextadd = sext i32 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we do not promote when the related instruction does not have
|
|
; the nsw flag.
|
|
; CHECK-LABEL: @noPromotion
|
|
; CHECK-NOT: add i64
|
|
; CHECK: ret
|
|
define i8 @noPromotion(i32 %arg1, i32 %arg2, i8* %base) {
|
|
%add = add i32 %arg1, %arg2
|
|
%sextadd = sext i32 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we correctly promote constant arguments.
|
|
; CHECK-LABEL: @oneArgPromotion
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i32 %arg1 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotion(i32 %arg1, i8* %base) {
|
|
%add = add nsw i32 %arg1, 1
|
|
%sextadd = sext i32 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we do not promote truncate when we cannot determine the
|
|
; bits that are dropped.
|
|
; CHECK-LABEL: @oneArgPromotionBlockTrunc1
|
|
; CHECK: [[ARG1TRUNC:%[a-zA-Z_0-9-]+]] = trunc i32 %arg1 to i8
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i8 [[ARG1TRUNC]] to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionBlockTrunc1(i32 %arg1, i8* %base) {
|
|
%trunc = trunc i32 %arg1 to i8
|
|
%add = add nsw i8 %trunc, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we do not promote truncate when we cannot determine all the
|
|
; bits that are dropped.
|
|
; CHECK-LABEL: @oneArgPromotionBlockTrunc2
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i16 %arg1 to i32
|
|
; CHECK: [[ARG1TRUNC:%[a-zA-Z_0-9-]+]] = trunc i32 [[ARG1SEXT]] to i8
|
|
; CHECK: [[ARG1SEXT64:%[a-zA-Z_0-9-]+]] = sext i8 [[ARG1TRUNC]] to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT64]], 1
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionBlockTrunc2(i16 %arg1, i8* %base) {
|
|
%sextarg1 = sext i16 %arg1 to i32
|
|
%trunc = trunc i32 %sextarg1 to i8
|
|
%add = add nsw i8 %trunc, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Check that we are able to promote truncate when we know all the bits
|
|
; that are dropped.
|
|
; CHECK-LABEL: @oneArgPromotionPassTruncKeepSExt
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i1 %arg1 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionPassTruncKeepSExt(i1 %arg1, i8* %base) {
|
|
%sextarg1 = sext i1 %arg1 to i32
|
|
%trunc = trunc i32 %sextarg1 to i8
|
|
%add = add nsw i8 %trunc, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; On X86 truncate are free. Check that we are able to promote the add
|
|
; to be used as addressing mode and that we insert a truncate for the other
|
|
; use.
|
|
; CHECK-LABEL: @oneArgPromotionTruncInsert
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i8 %arg1 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: [[TRUNC:%[a-zA-Z_0-9-]+]] = trunc i64 [[PROMOTED]] to i8
|
|
; CHECK: [[GEP:%[a-zA-Z_0-9-]+]] = getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: [[LOAD:%[a-zA-Z_0-9-]+]] = load i8* [[GEP]]
|
|
; CHECK: add i8 [[LOAD]], [[TRUNC]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionTruncInsert(i8 %arg1, i8* %base) {
|
|
%add = add nsw i8 %arg1, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
%finalres = add i8 %res, %add
|
|
ret i8 %finalres
|
|
}
|
|
|
|
; Cannot sext from a larger type than the promoted type.
|
|
; CHECK-LABEL: @oneArgPromotionLargerType
|
|
; CHECK: [[ARG1TRUNC:%[a-zA-Z_0-9-]+]] = trunc i128 %arg1 to i8
|
|
; CHECK: [[ARG1SEXT64:%[a-zA-Z_0-9-]+]] = sext i8 [[ARG1TRUNC]] to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT64]], 1
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionLargerType(i128 %arg1, i8* %base) {
|
|
%trunc = trunc i128 %arg1 to i8
|
|
%add = add nsw i8 %trunc, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
%finalres = add i8 %res, %add
|
|
ret i8 %finalres
|
|
}
|
|
|
|
; Use same inserted trunc
|
|
; On X86 truncate are free. Check that we are able to promote the add
|
|
; to be used as addressing mode and that we insert a truncate for
|
|
; *all* the other uses.
|
|
; CHECK-LABEL: @oneArgPromotionTruncInsertSeveralUse
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i8 %arg1 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: [[TRUNC:%[a-zA-Z_0-9-]+]] = trunc i64 [[PROMOTED]] to i8
|
|
; CHECK: [[GEP:%[a-zA-Z_0-9-]+]] = getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: [[LOAD:%[a-zA-Z_0-9-]+]] = load i8* [[GEP]]
|
|
; CHECK: [[ADDRES:%[a-zA-Z_0-9-]+]] = add i8 [[LOAD]], [[TRUNC]]
|
|
; CHECK: add i8 [[ADDRES]], [[TRUNC]]
|
|
; CHECK: ret
|
|
define i8 @oneArgPromotionTruncInsertSeveralUse(i8 %arg1, i8* %base) {
|
|
%add = add nsw i8 %arg1, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
%almostfinalres = add i8 %res, %add
|
|
%finalres = add i8 %almostfinalres, %add
|
|
ret i8 %finalres
|
|
}
|
|
|
|
; Check that the promoted instruction is used for all uses of the original
|
|
; sign extension.
|
|
; CHECK-LABEL: @oneArgPromotionSExtSeveralUse
|
|
; CHECK: [[ARG1SEXT:%[a-zA-Z_0-9-]+]] = sext i8 %arg1 to i64
|
|
; CHECK: [[PROMOTED:%[a-zA-Z_0-9-]+]] = add nsw i64 [[ARG1SEXT]], 1
|
|
; CHECK: [[GEP:%[a-zA-Z_0-9-]+]] = getelementptr inbounds i8* %base, i64 [[PROMOTED]]
|
|
; CHECK: [[LOAD:%[a-zA-Z_0-9-]+]] = load i8* [[GEP]]
|
|
; CHECK: [[ADDRES:%[a-zA-Z_0-9-]+]] = zext i8 [[LOAD]] to i64
|
|
; CHECK: add i64 [[ADDRES]], [[PROMOTED]]
|
|
; CHECK: ret
|
|
define i64 @oneArgPromotionSExtSeveralUse(i8 %arg1, i8* %base) {
|
|
%add = add nsw i8 %arg1, 1
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
%almostfinalres = zext i8 %res to i64
|
|
%finalres = add i64 %almostfinalres, %sextadd
|
|
ret i64 %finalres
|
|
}
|
|
|
|
; Check all types of rollback mechanism.
|
|
; For this test, the sign extension stays in place.
|
|
; However, the matching process goes until promoting both the operands
|
|
; of the first promotable add implies.
|
|
; At this point the rollback mechanism kicks in and restores the states
|
|
; until the addressing mode matcher is able to match something: in that
|
|
; case promote nothing.
|
|
; Along the way, the promotion mechanism involves:
|
|
; - Mutating the type of %promotableadd1 and %promotableadd2.
|
|
; - Creating a sext for %arg1 and %arg2.
|
|
; - Creating a trunc for a use of %promotableadd1.
|
|
; - Replacing a bunch of uses.
|
|
; - Setting the operands of the promoted instruction with the promoted values.
|
|
; - Moving instruction around (mainly sext when promoting instruction).
|
|
; Each type of those promotions has to be undo at least once during this
|
|
; specific test.
|
|
; CHECK-LABEL: @twoArgsPromotionNest
|
|
; CHECK: [[ORIG:%[a-zA-Z_0-9-]+]] = add nsw i32 %arg1, %arg2
|
|
; CHECK: [[ADD:%[a-zA-Z_0-9-]+]] = add nsw i32 [[ORIG]], [[ORIG]]
|
|
; CHECK: [[SEXT:%[a-zA-Z_0-9-]+]] = sext i32 [[ADD]] to i64
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[SEXT]]
|
|
; CHECK: ret
|
|
define i8 @twoArgsPromotionNest(i32 %arg1, i32 %arg2, i8* %base) {
|
|
%promotableadd1 = add nsw i32 %arg1, %arg2
|
|
%promotableadd2 = add nsw i32 %promotableadd1, %promotableadd1
|
|
%sextadd = sext i32 %promotableadd2 to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Test the InstructionRemover undo, which was the only one not
|
|
; kicked in the previous test.
|
|
; The matcher first promotes the add, removes the trunc and promotes
|
|
; the sext of arg1.
|
|
; Then, the matcher cannot use an addressing mode r + r + r, thus it
|
|
; rolls back.
|
|
; CHECK-LABEL: @twoArgsNoPromotionRemove
|
|
; CHECK: [[SEXTARG1:%[a-zA-Z_0-9-]+]] = sext i1 %arg1 to i32
|
|
; CHECK: [[TRUNC:%[a-zA-Z_0-9-]+]] = trunc i32 [[SEXTARG1]] to i8
|
|
; CHECK: [[ADD:%[a-zA-Z_0-9-]+]] = add nsw i8 [[TRUNC]], %arg2
|
|
; CHECK: [[SEXT:%[a-zA-Z_0-9-]+]] = sext i8 [[ADD]] to i64
|
|
; CHECK: getelementptr inbounds i8* %base, i64 [[SEXT]]
|
|
; CHECK: ret
|
|
define i8 @twoArgsNoPromotionRemove(i1 %arg1, i8 %arg2, i8* %base) {
|
|
%sextarg1 = sext i1 %arg1 to i32
|
|
%trunc = trunc i32 %sextarg1 to i8
|
|
%add = add nsw i8 %trunc, %arg2
|
|
%sextadd = sext i8 %add to i64
|
|
%arrayidx = getelementptr inbounds i8* %base, i64 %sextadd
|
|
%res = load i8* %arrayidx
|
|
ret i8 %res
|
|
}
|
|
|
|
; Ensure that when the profitability checks kicks in, the IR is not modified
|
|
; will IgnoreProfitability is on.
|
|
; The profitabily check happens when a candidate instruction has several uses.
|
|
; The matcher will create a new matcher for each use and check if the
|
|
; instruction is in the list of the matched instructions of this new matcher.
|
|
; All changes made by the new matchers must be dropped before pursuing
|
|
; otherwise the state of the original matcher will be wrong.
|
|
;
|
|
; Without the profitability check, when checking for the second use of
|
|
; arrayidx, the matcher promotes everything all the way to %arg1, %arg2.
|
|
; Check that we did not promote anything in the final matching.
|
|
;
|
|
; <rdar://problem/16020230>
|
|
; CHECK-LABEL: @checkProfitability
|
|
; CHECK-NOT: {{%[a-zA-Z_0-9-]+}} = sext i32 %arg1 to i64
|
|
; CHECK-NOT: {{%[a-zA-Z_0-9-]+}} = sext i32 %arg2 to i64
|
|
; CHECK: [[SHL:%[a-zA-Z_0-9-]+]] = shl nsw i32 %arg1, 1
|
|
; CHECK: [[ADD:%[a-zA-Z_0-9-]+]] = add nsw i32 [[SHL]], %arg2
|
|
; CHECK: [[SEXTADD:%[a-zA-Z_0-9-]+]] = sext i32 [[ADD]] to i64
|
|
; BB then
|
|
; CHECK: [[BASE1:%[a-zA-Z_0-9-]+]] = add i64 [[SEXTADD]], 48
|
|
; CHECK: [[ADDR1:%[a-zA-Z_0-9-]+]] = inttoptr i64 [[BASE1]] to i32*
|
|
; CHECK: load i32* [[ADDR1]]
|
|
; BB else
|
|
; CHECK: [[BASE2:%[a-zA-Z_0-9-]+]] = add i64 [[SEXTADD]], 48
|
|
; CHECK: [[ADDR2:%[a-zA-Z_0-9-]+]] = inttoptr i64 [[BASE2]] to i32*
|
|
; CHECK: load i32* [[ADDR2]]
|
|
; CHECK: ret
|
|
; CHECK-GEP-LABEL: @checkProfitability
|
|
; CHECK-GEP-NOT: {{%[a-zA-Z_0-9-]+}} = sext i32 %arg1 to i64
|
|
; CHECK-GEP-NOT: {{%[a-zA-Z_0-9-]+}} = sext i32 %arg2 to i64
|
|
; CHECK-GEP: [[SHL:%[a-zA-Z_0-9-]+]] = shl nsw i32 %arg1, 1
|
|
; CHECK-GEP: [[ADD:%[a-zA-Z_0-9-]+]] = add nsw i32 [[SHL]], %arg2
|
|
; CHECK-GEP: [[SEXTADD:%[a-zA-Z_0-9-]+]] = sext i32 [[ADD]] to i64
|
|
; BB then
|
|
; CHECK-GEP: [[BASE1:%[a-zA-Z_0-9-]+]] = inttoptr i64 [[SEXTADD]] to i32*
|
|
; CHECK-GEP: [[BCC1:%[a-zA-Z_0-9-]+]] = bitcast i32* [[BASE1]] to i8*
|
|
; CHECK-GEP: [[FULL1:%[a-zA-Z_0-9-]+]] = getelementptr i8* [[BCC1]], i64 48
|
|
; CHECK-GEP: [[ADDR1:%[a-zA-Z_0-9-]+]] = bitcast i8* [[FULL1]] to i32*
|
|
; CHECK-GEP: load i32* [[ADDR1]]
|
|
; BB else
|
|
; CHECK-GEP: [[BASE2:%[a-zA-Z_0-9-]+]] = inttoptr i64 [[SEXTADD]] to i32*
|
|
; CHECK-GEP: [[BCC2:%[a-zA-Z_0-9-]+]] = bitcast i32* [[BASE2]] to i8*
|
|
; CHECK-GEP: [[FULL2:%[a-zA-Z_0-9-]+]] = getelementptr i8* [[BCC2]], i64 48
|
|
; CHECK-GEP: [[ADDR2:%[a-zA-Z_0-9-]+]] = bitcast i8* [[FULL2]] to i32*
|
|
; CHECK-GEP: load i32* [[ADDR2]]
|
|
; CHECK-GEP: ret
|
|
define i32 @checkProfitability(i32 %arg1, i32 %arg2, i1 %test) {
|
|
%shl = shl nsw i32 %arg1, 1
|
|
%add1 = add nsw i32 %shl, %arg2
|
|
%sextidx1 = sext i32 %add1 to i64
|
|
%tmpptr = inttoptr i64 %sextidx1 to i32*
|
|
%arrayidx1 = getelementptr i32* %tmpptr, i64 12
|
|
br i1 %test, label %then, label %else
|
|
then:
|
|
%res1 = load i32* %arrayidx1
|
|
br label %end
|
|
else:
|
|
%res2 = load i32* %arrayidx1
|
|
br label %end
|
|
end:
|
|
%tmp = phi i32 [%res1, %then], [%res2, %else]
|
|
%res = add i32 %tmp, %add1
|
|
%addr = inttoptr i32 %res to i32*
|
|
%final = load i32* %addr
|
|
ret i32 %final
|
|
}
|