Re: [PATCH] D69897: Add #pragma clang loop vectorize_assume_alignment(n)
Hello Michael, A very good Morning to you.
On Wed, Nov 20, 2019 at 10:58 PM Michael Kruse wrote:
> Am Mi., 20. Nov. 2019 um 10:21 Uhr schrieb HAPPY Mahto
> :
> >> #pragma clang loop vectorize_assume_alignment(32)
> >> for(int i = 0;i < n; i++){
> >> a[i] = b[i] + i*i;
> >> }
> >
> > for this all-access inside the loop will be aligned to 32bit,
> > ex IR
> >>
> >> for.cond: ; preds = %for.inc,
> %entry
> >> %5 = load i32, i32* %i, align 32, !llvm.access.group !2
> >> %6 = load i32, i32* %n, align 32, !llvm.access.group !2
> >> %cmp = icmp slt i32 %5, %6
> >> br i1 %cmp, label %for.body, label %for.end
> >>
> >> for.body: ; preds = %for.cond
> >> %7 = load i32, i32* %i, align 32, !llvm.access.group !2
> >> %8 = load i32, i32* %i, align 32, !llvm.access.group !2
> >> %idxprom = sext i32 %8 to i64
> >> %arrayidx = getelementptr inbounds i32, i32* %vla1, i64 %idxprom
> >> store i32 %7, i32* %arrayidx, align 32, !llvm.access.group !2
> >> br label %for.inc
> >>
> >> for.inc: ; preds = %for.body
> >> %9 = load i32, i32* %i, align 32, !llvm.access.group !2
> >> %inc = add nsw i32 %9, 1
> >> store i32 %inc, i32* %i, align 32, !llvm.access.group !2
> >> br label %for.cond, !llvm.loop !3
> >
> > You will not need to create pointers for every array(or operand you want
> to perform the operation on).
>
> IMHO it is better if the programmer has to. It is not always obvious
> which arrays are used in the loop. Also, the information can be used
> by other optimzations that the vectorizer.
>
> We wrote this pragma in by keeping in mind that arrays inside the
important loops are used for vectorization and it'll be good for them to be
in aligned manner.
for( int i = 0; i < n; i++){
a[i] = b[i] + c[i];
d[i] = e[i] + i*i;
}
If there are more than 4-5 arrays inside loop being used then it'll be
extra effort to define all of them as _builtin_assume_aligned, we're
thinking of letting that thing handled by pragma itself. It'll be very
helpful for us if other people from community can give their views on this.
> >>
> >> void mult(float* x, int size, float factor){
> >> float* ax = (float*)__builtin_assume_aligned(x, 64);
> >> for (int i = 0; i < size; ++i)
> >> ax[i] *= factor;
> >> }
>
> https://godbolt.org/z/Fd6HMe
>
> > the alignment is assumed whereas in #pragma it is set to the number
> specified.
>
> Semantically, it is the same.
>
> I wonder how you expect the assembly output to change? The
> __builtin_assume_aligned, will be picked up by the backend and result
> in movaps to be used instead of movups.
>
>
> > it'll be easier, and having a pragma for doing this will help as it's
> provided in OMP and intel compilers.
>
> This is a compiler-specific extension. It does not have an influence
> on what other compilers do. Even with clang, if you try to do
>
> #pragma clang loop vectorize_assume_alignment(32)
> #pragma omp simd
> for (int i = 0; i < size; ++i)
>
> clang will silently swallow the vectorize_assume_alignment.
>
>
> Michael
>
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D69897: Add #pragma clang loop vectorize_assume_alignment(n)
Hello Michael,
Very sorry for the late reply, we had exams and assignments this week and I
had to read about _builtin_assume_aligned as I didn't come across this.
#pragma clang loop vectorize_assume_alignment(32)
> for(int i = 0;i < n; i++){
> a[i] = b[i] + i*i;
> }
>
for this all-access inside the loop will be aligned to 32bit,
ex IR
> for.cond: ; preds = %for.inc,
> %entry
> %5 = load i32, i32* %i, align 32, !llvm.access.group !2
> %6 = load i32, i32* %n, align 32, !llvm.access.group !2
> %cmp = icmp slt i32 %5, %6
> br i1 %cmp, label %for.body, label %for.end
>
> for.body: ; preds = %for.cond
> %7 = load i32, i32* %i, align 32, !llvm.access.group !2
> %8 = load i32, i32* %i, align 32, !llvm.access.group !2
> %idxprom = sext i32 %8 to i64
> %arrayidx = getelementptr inbounds i32, i32* %vla1, i64 %idxprom
> store i32 %7, i32* %arrayidx, align 32, !llvm.access.group !2
> br label %for.inc
>
> for.inc: ; preds = %for.body
> %9 = load i32, i32* %i, align 32, !llvm.access.group !2
> %inc = add nsw i32 %9, 1
> store i32 %inc, i32* %i, align 32, !llvm.access.group !2
> br label %for.cond, !llvm.loop !3
>
You will not need to create pointers for every array(or operand you want to
perform the operation on).
> void mult(float* x, int size, float factor){
> float* ax = (float*)__builtin_assume_aligned(x, 64);
> for (int i = 0; i < size; ++i)
> ax[i] *= factor;
> }
>
the IR generated for this :
> define void @mult(i32*, i32, float) #0 {
> %4 = alloca i32*, align 8
> %5 = alloca i32, align 4
> %6 = alloca float, align 4
> %7 = alloca i32*, align 8
> %8 = alloca i32, align 4
> store i32* %0, i32** %4, align 8
> store i32 %1, i32* %5, align 4
> store float %2, float* %6, align 4
> %9 = load i32*, i32** %4, align 8
> %10 = bitcast i32* %9 to i8*
> %11 = ptrtoint i8* %10 to i64
> %12 = and i64 %11, 63
> %13 = icmp eq i64 %12, 0
> call void @llvm.assume(i1 %13)
> %14 = bitcast i8* %10 to i32*
> store i32* %14, i32** %7, align 8
> store i32 0, i32* %8, align 4
> br label %15
>
> ; :15: ; preds = %29, %3
> %16 = load i32, i32* %8, align 4
> %17 = load i32, i32* %5, align 4
> %18 = icmp slt i32 %16, %17
> br i1 %18, label %19, label %32
>
> ; :19: ; preds = %15
> %20 = load float, float* %6, align 4
> %21 = load i32*, i32** %7, align 8
> %22 = load i32, i32* %8, align 4
> %23 = sext i32 %22 to i64
> %24 = getelementptr inbounds i32, i32* %21, i64 %23
> %25 = load i32, i32* %24, align 4
> %26 = sitofp i32 %25 to float
> %27 = fmul float %26, %20
> %28 = fptosi float %27 to i32
> store i32 %28, i32* %24, align 4
> br label %29
>
> ; :29: ; preds = %19
> %30 = load i32, i32* %8, align 4
> %31 = add nsw i32 %30, 1
> store i32 %31, i32* %8, align 4
> br label %15
>
> ; :32: ; preds = %15
> ret void
> }
>
the alignment is assumed whereas in #pragma it is set to the number
specified.
it'll be easier, and having a pragma for doing this will help as it's
provided in OMP and intel compilers.
Thank you, If I made any mistake please tell me.
Happy Mahto
CSE Undergrad, IIT Hyderabad
On Thu, Nov 14, 2019 at 10:32 PM Michael Kruse via Phabricator <
revi...@reviews.llvm.org> wrote:
> Meinersbur added a comment.
>
> Could you elaborate why this is better than `__builtin_assume_aligned`?
>
>
> Repository:
> rG LLVM Github Monorepo
>
> CHANGES SINCE LAST ACTION
> https://reviews.llvm.org/D69897/new/
>
> https://reviews.llvm.org/D69897
>
>
>
>
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D69897: Add #pragma clang loop aligned
On Wed, Nov 6, 2019 at 10:43 PM Hal Finkel via Phabricator < revi...@reviews.llvm.org> wrote: > hfinkel added inline comments. > > > > Comment at: clang/docs/LanguageExtensions.rst:3135 > + > +This predicates all the array references inside the loop to be aligned. > The aligned access to them can increase fetch time and increase the > performance. > + > > lebedev.ri wrote: > > What does this actually mean, in the end? > > Will it assume that whatever alignment is required > > for whatever vectorization width chosen, > > is the actual alignment? > Currently, it is using 32 bit. > Also, just arrays, or also pointers? > > By pointers, if you mean pointers to arrays? Then Yes. ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
