Hi all,
I'm having trouble understanding how llvm determines if pointers
alias. Consider the following two functions that each do a redundant
load:
define float @A(float * noalias %ptr1) {
%ptr2 = getelementptr float* %ptr1, i32 1024
%val1a = load float* %ptr1
store float %val1a, float* %ptr2
%val1b = load float* %ptr1
ret float %val1b
}
define float @B(float * noalias %ptr1, float * noalias %ptr2) {
%val1a = load float* %ptr1
store float %val1a, float* %ptr2
%val1b = load float* %ptr1
ret float %val1b
}
When I throw this code into test_alias.ll and run:
opt test_alias.ll -basicaa -print-alias-sets > /dev/null
it tells me that in the first function, ptr1 and ptr2 may alias, and
in the second function they do not. The source of
BasicAliasAnalysis.cpp seems more than intelligent enough to detect
both of these cases as not aliasing.
More to the point, the X86 assembly I get from compiling this is:
opt -O3 test_alias.ll | llc -filetype=asm
...
_A: ## @A
movss (%rdi), %xmm0
movss %xmm0, 4096(%rdi)
movss (%rdi), %xmm0
ret
...
_B: ## @B
movss (%rdi), %xmm0
movss %xmm0, (%rsi)
ret
...
Note the redundant movss in A.
Any idea what I'm doing wrong?
I am using what I believe is the most recent svn version (llvm version
3.0svn, revision 141415).
- Andrew
On Fri, Oct 7, 2011 at 2:15 PM, andrew adams <andrew.b.adams at gmail.com> wrote:> Hi all, > > I'm having trouble understanding how llvm determines if pointers > alias. Consider the following two functions that each do a redundant > load: > > define float @A(float * noalias %ptr1) { > %ptr2 = getelementptr float* %ptr1, i32 1024 > %val1a = load float* %ptr1 > store float %val1a, float* %ptr2 > %val1b = load float* %ptr1 > ret float %val1b > } > > > define float @B(float * noalias %ptr1, float * noalias %ptr2) { > %val1a = load float* %ptr1 > store float %val1a, float* %ptr2 > %val1b = load float* %ptr1 > ret float %val1b > } > > When I throw this code into test_alias.ll and run: > > opt test_alias.ll -basicaa -print-alias-sets > /dev/null > > it tells me that in the first function, ptr1 and ptr2 may alias, and > in the second function they do not. The source of > BasicAliasAnalysis.cpp seems more than intelligent enough to detect > both of these cases as not aliasing. > > More to the point, the X86 assembly I get from compiling this is: > > opt -O3 test_alias.ll | llc -filetype=asm > ... > _A: ## @A > movss (%rdi), %xmm0 > movss %xmm0, 4096(%rdi) > movss (%rdi), %xmm0 > ret > > ... > _B: ## @B > movss (%rdi), %xmm0 > movss %xmm0, (%rsi) > ret > ... > > Note the redundant movss in A. > > Any idea what I'm doing wrong?Are you providing target data? AA skips a bunch of calculations when the the size of a pointer isn't specified. -Eli
On Fri, Oct 7, 2011 at 5:43 PM, Eli Friedman <eli.friedman at gmail.com> wrote:> On Fri, Oct 7, 2011 at 2:15 PM, andrew adams <andrew.b.adams at gmail.com> wrote: >> Hi all, >> >> I'm having trouble understanding how llvm determines if pointers >> alias. Consider the following two functions that each do a redundant >> load: >> >> define float @A(float * noalias %ptr1) { >> %ptr2 = getelementptr float* %ptr1, i32 1024 >> %val1a = load float* %ptr1 >> store float %val1a, float* %ptr2 >> %val1b = load float* %ptr1 >> ret float %val1b >> } >> >> >> define float @B(float * noalias %ptr1, float * noalias %ptr2) { >> %val1a = load float* %ptr1 >> store float %val1a, float* %ptr2 >> %val1b = load float* %ptr1 >> ret float %val1b >> } >> >> When I throw this code into test_alias.ll and run: >> >> opt test_alias.ll -basicaa -print-alias-sets > /dev/null >> >> it tells me that in the first function, ptr1 and ptr2 may alias, and >> in the second function they do not. The source of >> BasicAliasAnalysis.cpp seems more than intelligent enough to detect >> both of these cases as not aliasing. >> >> More to the point, the X86 assembly I get from compiling this is: >> >> opt -O3 test_alias.ll | llc -filetype=asm >> ... >> _A: ## @A >> movss (%rdi), %xmm0 >> movss %xmm0, 4096(%rdi) >> movss (%rdi), %xmm0 >> ret >> >> ... >> _B: ## @B >> movss (%rdi), %xmm0 >> movss %xmm0, (%rsi) >> ret >> ... >> >> Note the redundant movss in A. >> >> Any idea what I'm doing wrong? > > Are you providing target data? AA skips a bunch of calculations when > the the size of a pointer isn't specified. > > -Eli >Thanks, that was it. - Andrew