On Tue, May 25, 2021 at 4:29 PM Richard Biener
<richard.guent...@gmail.com> wrote:
>
> On Fri, May 21, 2021 at 5:00 PM Uros Bizjak via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
> >
> > Here it is, the patch that enables auto-vectorization for 32bit modes.
> >
> > Sent as RFC, because the patch fails some vectorizer scans, as it
> > obviously enables more vectorization to happen:
> >
> > Running target unix
> > FAIL: gcc.dg/vect/pr71264.c -flto -ffat-lto-objects scan-tree-dump
> > vect "vectorized 1 loops in function"
> > FAIL: gcc.dg/vect/pr71264.c scan-tree-dump vect "vectorized 1 loops in
> > function"
> > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorized 1 loops" 1
> > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorizing stmts using SLP" 1
> > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorized 1 loops" 1
> > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorizing
> > stmts using SLP" 1
> > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorized 3 loops" 1
> > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorizing stmts using SLP" 3
> > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorized 3 loops" 1
> > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorizing stmts
> > using SLP" 3
> >
> >
> > Running target unix/-m32
> > FAIL: gcc.dg/vect/no-vfa-vect-101.c scan-tree-dump-times vect "can't
> > determine dependence" 1
> > FAIL: gcc.dg/vect/no-vfa-vect-102.c scan-tree-dump-times vect
> > "possible dependence between data-refs" 1
> > FAIL: gcc.dg/vect/no-vfa-vect-102a.c scan-tree-dump-times vect
> > "possible dependence between data-refs" 1
> > FAIL: gcc.dg/vect/no-vfa-vect-37.c scan-tree-dump-times vect "can't
> > determine dependence" 2
> > FAIL: gcc.dg/vect/pr71264.c -flto -ffat-lto-objects scan-tree-dump
> > vect "vectorized 1 loops in function"
> > FAIL: gcc.dg/vect/pr71264.c scan-tree-dump vect "vectorized 1 loops in
> > function"
> > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorized 1 loops" 1
> > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorizing stmts using SLP" 1
> > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorized 1 loops" 1
> > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorizing
> > stmts using SLP" 1
> > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorized 3 loops" 1
> > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "vectorizing stmts using SLP" 3
> > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorized 3 loops" 1
> > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorizing stmts
> > using SLP" 3
> > FAIL: gcc.dg/vect/vect-104.c -flto -ffat-lto-objects
> > scan-tree-dump-times vect "possible dependence between data-refs" 1
> > FAIL: gcc.dg/vect/vect-104.c scan-tree-dump-times vect "possible
> > dependence between data-refs" 1
>
> Yeah, it's a bit iffy to adjust expectations. If there's a way to
> disable vectorization
> for 32bit modes on x86 that might be a way to "fix" them, otherwise we're
> lacking a way to query for available vector modes/sizes in the dejagnu vect
> targets. There's available_vector_sizes but it's implementation is hardly
> complete nor is size the only important thing (FP vs. INT). At least
> one could add a vect32 predicate similar to the existing vect64 one.
I went the way you proposed above. By adding 32bit vector size to
available_vector_sizes only two testcases fails. The attached patch
fixes all vect scan failures (the remaining failure in
vect_epilogues.c is just the case of missing uavg<mode>3_ceil pattern
for V4QI epilogue vectorization - I plan to add the insn in the
follow-up patch).
The patch also xfails pr71264.c, the case of missing re-vectorization
of 32bit vectors.
WDYT?
Uros.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 28e6113a609..04649b42122 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -22190,12 +22190,15 @@ ix86_autovectorize_vector_modes (vector_modes *modes,
bool all)
modes->safe_push (V16QImode);
modes->safe_push (V32QImode);
}
- else if (TARGET_MMX_WITH_SSE)
+ else if (TARGET_SSE2)
modes->safe_push (V16QImode);
if (TARGET_MMX_WITH_SSE)
modes->safe_push (V8QImode);
+ if (TARGET_SSE2)
+ modes->safe_push (V4QImode);
+
return 0;
}
diff --git a/gcc/doc/sourcebuild.texi b/gcc/doc/sourcebuild.texi
index cf3098749c0..16c6a3b8e99 100644
--- a/gcc/doc/sourcebuild.texi
+++ b/gcc/doc/sourcebuild.texi
@@ -1740,6 +1740,12 @@ circumstances.
@item vect_variable_length
Target has variable-length vectors.
+@item vect64
+Target supports vectors of 64 bits.
+
+@item vect32
+Target supports vectors of 32 bits.
+
@item vect_widen_sum_hi_to_si
Target supports a vector widening summation of @code{short} operands
into @code{int} results, or can promote (unpack) from @code{short}
diff --git a/gcc/testsuite/gcc.dg/vect/pr71264.c
b/gcc/testsuite/gcc.dg/vect/pr71264.c
index dc849bf2797..1381e0ed132 100644
--- a/gcc/testsuite/gcc.dg/vect/pr71264.c
+++ b/gcc/testsuite/gcc.dg/vect/pr71264.c
@@ -19,5 +19,4 @@ void test(uint8_t *ptr, uint8_t *mask)
}
}
-/* { dg-final { scan-tree-dump "vectorized 1 loops in function" "vect" { xfail
s390*-*-* sparc*-*-* } } } */
-
+/* { dg-final { scan-tree-dump "vectorized 1 loops in function" "vect" { xfail
{ { s390*-*-* sparc*-*-* } || vect32 } } } } */
diff --git a/gcc/testsuite/gcc.dg/vect/slp-28.c
b/gcc/testsuite/gcc.dg/vect/slp-28.c
index 7778bad4465..0bb5f0eb0e4 100644
--- a/gcc/testsuite/gcc.dg/vect/slp-28.c
+++ b/gcc/testsuite/gcc.dg/vect/slp-28.c
@@ -88,6 +88,7 @@ int main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" }
} */
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target {
! vect32 } } } } */
+/* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { target
vect32 } } } */
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" {
target { ! vect32 } } } } */
diff --git a/gcc/testsuite/gcc.dg/vect/slp-3.c
b/gcc/testsuite/gcc.dg/vect/slp-3.c
index 46ab584419a..80ded1840ad 100644
--- a/gcc/testsuite/gcc.dg/vect/slp-3.c
+++ b/gcc/testsuite/gcc.dg/vect/slp-3.c
@@ -141,8 +141,8 @@ int main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { target {
! vect_partial_vectors } } } } */
-/* { dg-final { scan-tree-dump-times "vectorized 4 loops" 1 "vect" { target
vect_partial_vectors } } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" {
target { ! vect_partial_vectors } } } }*/
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 4 "vect" {
target vect_partial_vectors } } } */
+/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { target {
! { vect_partial_vectors || vect32 } } } } } */
+/* { dg-final { scan-tree-dump-times "vectorized 4 loops" 1 "vect" { target {
vect_partial_vectors || vect32 } } } } */
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" {
target { ! { vect_partial_vectors || vect32 } } } } }*/
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 4 "vect" {
target { vect_partial_vectors || vect32 } } } } */
diff --git a/gcc/testsuite/lib/target-supports.exp
b/gcc/testsuite/lib/target-supports.exp
index 849f1bbeda5..7f78c5593ac 100644
--- a/gcc/testsuite/lib/target-supports.exp
+++ b/gcc/testsuite/lib/target-supports.exp
@@ -7626,6 +7626,7 @@ proc available_vector_sizes { } {
if { ![is-effective-target ia32] } {
lappend result 64
}
+ lappend result 32
} elseif { [istarget sparc*-*-*] } {
lappend result 64
} elseif { [istarget amdgcn*-*-*] } {
@@ -7655,6 +7656,12 @@ proc check_effective_target_vect64 { } {
return [expr { [lsearch -exact [available_vector_sizes] 64] >= 0 }]
}
+# Return 1 if the target supports vectors of 32 bits.
+
+proc check_effective_target_vect32 { } {
+ return [expr { [lsearch -exact [available_vector_sizes] 32] >= 0 }]
+}
+
# Return 1 if the target supports vector copysignf calls.
proc check_effective_target_vect_call_copysignf { } {
