procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

middle-end: Handle FMA_CONJ correctly after SLP layout update.

Browse source 
Apologies, I got dinged by the i386 regressions bot for a test I didn't have in
my tree at the time I made the previous patch.  The bot was telling me that FMA
stopped working after I strengthened the FMA check in the previous patch.

The reason is that the check is slightly early.  The first check can indeed only
exit early when either node isn't a mult.  However we need to delay till we know
if the node is a MUL or FMA before enforcing that both nodes must be a MULT
since the node to inspect is different if the operation is a MUL or FMA.

Also with the update patch for GCC 11 tree layout update to the new GCC 12 one
I had missed that the difference in which node is conjucated is not symmetrical.

So the test for it can just be testing the inverse order.  It was Currently
no detecting when the first node was conjucated instead of the second one.

This also made me wonder why my own test didn't detect this.  It turns out that
the tests, being copied from the _Float16 ones were incorrectly marked as
xfail.  The _Float16 ones are marked as xfail since C doesn't have a conj
operation for _Float16, which means you get extra type-casts in between.

While you could use the GCC _Complex extension here I opted to mark them xfail
since I wanted to include detection over the widenings next year.

Secondly the double tests were being skipped because Adv. SIMD was missing from
targets supporting Complex Double vectorization.

With these changes all other tests run and pass and only XFAIL ones are
correctly the _Float16 ones.  Sorry for missing this before, testing should now
cover all cases.

gcc/ChangeLog:

	PR tree-optimization/103311
	PR target/103330
	* tree-vect-slp-patterns.c (vect_validate_multiplication): Fix CONJ
	test to new codegen.
	(complex_mul_pattern::matches): Move check downwards.

gcc/testsuite/ChangeLog:

	PR tree-optimization/103311
	PR target/103330
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-double.c: Fix it.
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-float.c: Likewise.
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-double.c: Likewise.
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-float.c: Likewise.
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-double.c: Likewise.
	* gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-float.c: Likewise.
	* lib/target-supports.exp
	(check_effective_target_vect_complex_add_double): Add Adv. SIMD.
This commit is contained in:
Tamar Christina 2021-11-19 15:12:38 +00:00
parent 79e9f721d1
commit 487d604b6f
8 changed files with 37 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-double.c
View file

@ -1,10 +1,11 @@
/* { dg-do compile } */
/* { dg-require-effective-target vect_complex_add_double } */
/* { dg-add-options arm_v8_3a_complex_neon } */
/* { dg-additional-options "-fdump-tree-vect-details" } */
#define TYPE double
#define N 16
#include "complex-mla-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA_CONJ" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA_CONJ" "vect" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA" "vect" } } */

6
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-float.c
View file

@ -1,10 +1,10 @@
/* { dg-do compile } */
/* { dg-require-effective-target vect_complex_add_float } */
/* { dg-additional-options "-fno-tree-loop-vectorize" } */
/* { dg-additional-options "-fdump-tree-vect-details" } */
/* { dg-add-options arm_v8_3a_fp16_complex_neon } */
#define TYPE float
#define N 16
#include "complex-mla-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA_CONJ" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA_CONJ" "vect" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA" "vect" } } */

7
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-double.c
View file

@ -1,12 +1,11 @@
/* { dg-do compile } */
/* { dg-require-effective-target vect_complex_add_double } */
/* { dg-add-options arm_v8_3a_complex_neon } */
/* { dg-additional-options "-fdump-tree-vect-details" } */
#define TYPE double
#define N 16
#include "complex-mls-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_ADD_ROT270" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMA" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS_CONJ" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS_CONJ" "vect" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS" "vect" } } */

6
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-float.c
View file

@ -1,11 +1,11 @@
/* { dg-do compile } */
/* { dg-require-effective-target vect_complex_add_float } */
/* { dg-additional-options "-fno-tree-loop-vectorize" } */
/* { dg-additional-options "-fdump-tree-vect-details" } */
/* { dg-add-options arm_v8_3a_complex_neon } */
#define TYPE float
#define N 16
#include "complex-mls-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS_CONJ" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS_CONJ" "vect" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_FMS" "vect" } } */

5
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-double.c
View file

@ -1,10 +1,11 @@
/* { dg-do compile } */
/* { dg-require-effective-target vect_complex_add_double } */
/* { dg-add-options arm_v8_3a_complex_neon } */
/* { dg-additional-options "-fdump-tree-vect-details" } */
#define TYPE double
#define N 16
#include "complex-mul-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL_CONJ" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL" slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL_CONJ" "vect" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL" "vect" } } */

4
gcc/testsuite/gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-float.c
View file

@ -7,5 +7,5 @@
#define N 16
#include "complex-mul-template.c"
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL_CONJ" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL" "slp1" { xfail *-*-* } } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL_CONJ" "slp1" } } */
/* { dg-final { scan-tree-dump "Found COMPLEX_MUL" "slp1" } } */

6
gcc/testsuite/lib/target-supports.exp
View file

@ -3632,8 +3632,10 @@ proc check_effective_target_vect_complex_add_float { } {
proc check_effective_target_vect_complex_add_double { } {
return [check_cached_effective_target_indexed vect_complex_add_double {
expr {
([check_effective_target_aarch64_sve2]
&& [check_effective_target_aarch64_little_endian])
(([check_effective_target_arm_v8_3a_complex_neon_ok]
&& [check_effective_target_aarch64_little_endian])
|| ([check_effective_target_aarch64_sve2]
&& [check_effective_target_aarch64_little_endian]))
}}]
}

24
gcc/tree-vect-slp-patterns.c
View file

@ -809,14 +809,20 @@ vect_validate_multiplication (slp_tree_to_load_perm_map_t *perm_cache,
if (linear_loads_p (perm_cache, left_op[index2]) == PERM_EVENODD)
return true;
}
else if (kind == PERM_EVENODD)
else if (kind == PERM_EVENODD && !neg_first)
{
if ((kind = linear_loads_p (perm_cache, left_op[index2])) == PERM_EVENODD)
if ((kind = linear_loads_p (perm_cache, left_op[index2])) != PERM_EVENEVEN)
return false;
return true;
}
else if (!neg_first)
*conj_first_operand = true;
else if (kind == PERM_EVENEVEN && neg_first)
{
if ((kind = linear_loads_p (perm_cache, left_op[index2])) != PERM_EVENODD)
return false;
*conj_first_operand = true;
return true;
}
else
return false;
@ -949,7 +955,7 @@ complex_mul_pattern::matches (complex_operation_t op,
bool mul0 = vect_match_expression_p (l0node[0], MULT_EXPR);
bool mul1 = vect_match_expression_p (l0node[1], MULT_EXPR);
if (!mul0 || !mul1)
if (!mul0 && !mul1)
return IFN_LAST;
/* Now operand2+4 may lead to another expression. */
@ -962,7 +968,7 @@ complex_mul_pattern::matches (complex_operation_t op,
{
auto vals = SLP_TREE_CHILDREN (l0node[0]);
/* Check if it's a multiply, otherwise no idea what this is. */
if (!vect_match_expression_p (vals[1], MULT_EXPR))
if (!(mul0 = vect_match_expression_p (vals[1], MULT_EXPR)))
return IFN_LAST;
/* Check if the ADD is linear, otherwise it's not valid complex FMA. */
@ -979,6 +985,8 @@ complex_mul_pattern::matches (complex_operation_t op,
if (left_op.length () != 2
|| right_op.length () != 2
|| !mul0
|| !mul1
|| linear_loads_p (perm_cache, left_op[1]) == PERM_ODDEVEN)
return IFN_LAST;
@ -993,7 +1001,7 @@ complex_mul_pattern::matches (complex_operation_t op,
if (!vect_validate_multiplication (perm_cache, left_op, PERM_EVENEVEN)
|| vect_normalize_conj_loc (left_op))
return IFN_LAST;
if (!mul0)
if (add0)
ifn = IFN_COMPLEX_FMA;
else
ifn = IFN_COMPLEX_MUL;
@ -1005,7 +1013,7 @@ complex_mul_pattern::matches (complex_operation_t op,
false))
return IFN_LAST;
if(!mul0)
if(add0)
ifn = IFN_COMPLEX_FMA_CONJ;
else
ifn = IFN_COMPLEX_MUL_CONJ;