On 05/19/2016 01:40 PM, Ilya Enkovich wrote:
Hi,
This patch extends vectorizer cost model to include masking cost by
adding new cost model locations and new target hook to compute
masking cost.
Thanks,
Ilya
--
gcc/
2016-05-19 Ilya Enkovich <ilya.enkov...@intel.com>
* config/i386/i386.c (ix86_init_cost): Extend costs array.
(ix86_add_stmt_masking_cost): New.
(ix86_finish_cost): Add masking_prologue_cost and masking_body_cost
args.
(TARGET_VECTORIZE_ADD_STMT_MASKING_COST): New.
* config/i386/i386.h (TARGET_INCREASE_MASK_STORE_COST): New.
* config/i386/x86-tune.def (X86_TUNE_INCREASE_MASK_STORE_COST): New.
* config/rs6000/rs6000.c (_rs6000_cost_data): Extend cost array.
(rs6000_init_cost): Initialize new cost elements.
(rs6000_finish_cost): Add masking_prologue_cost and masking_body_cost.
* config/spu/spu.c (spu_init_cost): Extend costs array.
(spu_finish_cost): Add masking_prologue_cost and masking_body_cost args.
* doc/tm.texi.in (TARGET_VECTORIZE_ADD_STMT_MASKING_COST): New.
* doc/tm.texi: Regenerated.
* target.def (add_stmt_masking_cost): New.
(finish_cost): Add masking_prologue_cost and masking_body_cost args.
* target.h (enum vect_cost_for_stmt): Add vector_mask_load and
vector_mask_store.
(enum vect_cost_model_location): Add vect_masking_prologue
and vect_masking_body.
* targhooks.c (default_builtin_vectorization_cost): Support
vector_mask_load and vector_mask_store.
(default_init_cost): Extend costs array.
(default_add_stmt_masking_cost): New.
(default_finish_cost): Add masking_prologue_cost and masking_body_cost
args.
* targhooks.h (default_add_stmt_masking_cost): New.
* tree-vect-loop.c (vect_estimate_min_profitable_iters): Adjust
finish_cost call.
* tree-vect-slp.c (vect_bb_vectorization_profitable_p): Likewise.
* tree-vectorizer.h (add_stmt_masking_cost): New.
(finish_cost): Add masking_prologue_cost and masking_body_cost args.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 9f62089..6c2c364 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -53932,8 +53932,12 @@ ix86_spill_class (reg_class_t rclass, machine_mode
mode)
static void *
ix86_init_cost (struct loop *)
{
- unsigned *cost = XNEWVEC (unsigned, 3);
- cost[vect_prologue] = cost[vect_body] = cost[vect_epilogue] = 0;
+ unsigned *cost = XNEWVEC (unsigned, 5);
+ cost[vect_prologue] = 0;
+ cost[vect_body] = 0;
+ cost[vect_epilogue] = 0;
+ cost[vect_masking_prologue] = 0;
+ cost[vect_masking_body] = 0;
return cost;
Trivial nit -- no need or desire to use whitespace to line up the
initializers. It looks like others may have done this in the
duplicated instances of finish_cost. But we shouldn't propagate that
mistake into the init_cost hooks ;-)
@@ -1115,8 +1117,12 @@ default_get_mask_mode (unsigned nunits, unsigned
vector_size)
void *
default_init_cost (struct loop *loop_info ATTRIBUTE_UNUSED)
{
- unsigned *cost = XNEWVEC (unsigned, 3);
- cost[vect_prologue] = cost[vect_body] = cost[vect_epilogue] = 0;
+ unsigned *cost = XNEWVEC (unsigned, 5);
+ cost[vect_prologue] = 0;
+ cost[vect_body] = 0;
+ cost[vect_epilogue] = 0;
+ cost[vect_masking_prologue] = 0;
+ cost[vect_masking_body] = 0;
return cost;
Here too. There's others. I won't point them all out. Please double
check for this nit in any added code. You don't have to go back and fix
existing problems of this nature.
I don't see anything here I really object to -- Richi and I may disagree
on the compute-costs once in a single scan vs restarting the scan. If
Richi feels strongly about restarting for some reason, I'll defer to him
-- he's done more work in the vectorizer than myself.
I'd suggest taking another stab at the docs for the hooks based on
Richi's question about whether or not the hook returns the cost of hte
masked statement or the cost of masking the statement.
jeff
