Flipping the order of the operands can end up changing the sign of the operation
and to compensate an adjustment operation has to be inserted (i.e. * {-1,1}).
The current code in trunk cannot do this yet so backporting the change from my
complex numbers branch caused it to produce incorrect code.
For now while the full changes aren't upstreamed remove the last 4 swaps.
Bootstrapped Regtested on aarch64-none-linux-gnu,
arm-none-linux-gnueabihf, x86_64-pc-linux-gnu
-m32, -m64 and no issues.
Pushed.
Thanks,
Tamar
gcc/ChangeLog:
* tree-vect-slp-patterns.cc (vect_validate_multiplication): Cleanup
comment.
(vect_validate_multiplication_commutative): Remove last 4 orders.
---
diff --git a/gcc/tree-vect-slp-patterns.cc b/gcc/tree-vect-slp-patterns.cc
index
aae000ee5292ccb1521928e43f514a14386a7827..8a03a25475da8d944a48da8a5ad88220d027f3f8
100644
--- a/gcc/tree-vect-slp-patterns.cc
+++ b/gcc/tree-vect-slp-patterns.cc
@@ -933,9 +933,8 @@ vect_validate_multiplication (slp_tree_to_load_perm_map_t
*perm_cache,
/* Try to validate LEFT_OP and RIGHT_OP as the operands of a complex
multiplication. Since MULT_EXPR is commutative, try all combinations of
- swapping the operands of each multiplication and both orders of the two
- multiplies. If a match is found, set OPS and STATUS for the matching
- order. */
+ swapping the operands of each multiplication. If a match is found, set OPS
+ and STATUS for the matching order. */
static inline bool
vect_validate_multiplication_commutative (slp_tree_to_load_perm_map_t
*perm_cache,
@@ -951,18 +950,13 @@ vect_validate_multiplication_commutative
(slp_tree_to_load_perm_map_t *perm_cach
{ 0, 1, 3, 2 }, /* (L0 * L1), (R1 * R0). */
{ 1, 0, 2, 3 }, /* (L1 * L0), (R0 * R1). */
{ 1, 0, 3, 2 }, /* (L1 * L0), (R1 * R0). */
- { 2, 3, 0, 1 }, /* (R0 * R1), (L0 * L1). */
- { 2, 3, 1, 0 }, /* (R0 * R1), (L1 * L0). */
- { 3, 2, 0, 1 }, /* (R1 * R0), (L0 * L1). */
- { 3, 2, 1, 0 }, /* (R1 * R0), (L1 * L0). */
};
- /* The first four entries only swap operands within each MULT_EXPR.
- The remaining entries also swap the two product terms, which is not
- valid for plain subtraction. */
- unsigned nperms = subtract ? 4 : ARRAY_SIZE (op_indices);
+ /* Only try permutations that swap operands within each MULT_EXPR. Swapping
+ the two product terms is not valid because the real lane is ordered by a
+ subtraction. */
slp_tree all_ops[4] = { left_op[0], left_op[1], right_op[0], right_op[1] };
- for (unsigned i = 0; i < nperms; ++i)
+ for (unsigned i = 0; i < ARRAY_SIZE (op_indices); ++i)
{
auto_vec<slp_tree> trial_ops;
if (vect_validate_multiplication (perm_cache, compat_cache, all_ops,
--
diff --git a/gcc/tree-vect-slp-patterns.cc b/gcc/tree-vect-slp-patterns.cc
index aae000ee5292ccb1521928e43f514a14386a7827..8a03a25475da8d944a48da8a5ad88220d027f3f8 100644
--- a/gcc/tree-vect-slp-patterns.cc
+++ b/gcc/tree-vect-slp-patterns.cc
@@ -933,9 +933,8 @@ vect_validate_multiplication (slp_tree_to_load_perm_map_t *perm_cache,
/* Try to validate LEFT_OP and RIGHT_OP as the operands of a complex
multiplication. Since MULT_EXPR is commutative, try all combinations of
- swapping the operands of each multiplication and both orders of the two
- multiplies. If a match is found, set OPS and STATUS for the matching
- order. */
+ swapping the operands of each multiplication. If a match is found, set OPS
+ and STATUS for the matching order. */
static inline bool
vect_validate_multiplication_commutative (slp_tree_to_load_perm_map_t *perm_cache,
@@ -951,18 +950,13 @@ vect_validate_multiplication_commutative (slp_tree_to_load_perm_map_t *perm_cach
{ 0, 1, 3, 2 }, /* (L0 * L1), (R1 * R0). */
{ 1, 0, 2, 3 }, /* (L1 * L0), (R0 * R1). */
{ 1, 0, 3, 2 }, /* (L1 * L0), (R1 * R0). */
- { 2, 3, 0, 1 }, /* (R0 * R1), (L0 * L1). */
- { 2, 3, 1, 0 }, /* (R0 * R1), (L1 * L0). */
- { 3, 2, 0, 1 }, /* (R1 * R0), (L0 * L1). */
- { 3, 2, 1, 0 }, /* (R1 * R0), (L1 * L0). */
};
- /* The first four entries only swap operands within each MULT_EXPR.
- The remaining entries also swap the two product terms, which is not
- valid for plain subtraction. */
- unsigned nperms = subtract ? 4 : ARRAY_SIZE (op_indices);
+ /* Only try permutations that swap operands within each MULT_EXPR. Swapping
+ the two product terms is not valid because the real lane is ordered by a
+ subtraction. */
slp_tree all_ops[4] = { left_op[0], left_op[1], right_op[0], right_op[1] };
- for (unsigned i = 0; i < nperms; ++i)
+ for (unsigned i = 0; i < ARRAY_SIZE (op_indices); ++i)
{
auto_vec<slp_tree> trial_ops;
if (vect_validate_multiplication (perm_cache, compat_cache, all_ops,
