| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Jan Sjodin <jan.sjodin@amd.com>
* tree-vect-analyze.c (vect_analyze_operations): Change
comparison of loop iterations with threshold to less than
or equal to instead of less than. Reduce
min_scalar_loop_bound by one.
* tree-vect-transform.c (vect_estimate_min_profitable_iters):
Change prologue and epilogue iterations estimate to vf/2,
when unknown at compile-time. Change versioning guard
cost to taken_branch_cost. If peeling for alignment is
unknown at compile-time, change peel guard costs to one
taken branch and one not-taken branch per peeled loop.
If peeling for alignment is known but number of scalar loop
iterations is unknown at compile-time, change peel guard
costs to one taken branch per peeled loop. Change the cost
model equation to consider vector iterations as the loop
iterations less the prologue and epilogue iterations.
Change outside vector cost check to less than or equal to
zero instead of equal to zero.
(vect_do_peeling_for_loop_bound): Reduce
min_scalar_loop_bound by one.
* tree-vectorizer.h: Add TARG_COND_TAKEN_BRANCH_COST and
TARG_COND_NOT_TAKEN_BRANCH_COST.
* config/i386/i386.h (processor_costs): Add
scalar_stmt_cost, scalar_load_cost, scalar_store_cost,
vec_stmt_cost, vec_to_scalar_cost, scalar_to_vec_cost,
vec_align_load_cost, vect_unalign_load_cost,
vec_store_cost, cond_taken_branch_cost,
cond_not_taken_branch_cost.
Define macros for x86 costs.
* config/i386/i386.c:
(size_cost): Set scalar_stmt_cost, scalar_load_cost,
scalar_store_cost, vec_stmt_cost, vec_to_scalar_cost,
scalar_to_vec_cost, vec_align_load_cost,
vect_unalign_load_cost, vec_store_cost,
cond_taken_branch_cost, cond_not_taken_branch_cost to one.
(i386_cost, i486_cost, pentium_cost, pentiumpro_cost,
geode_cost, k6_cost, athlon_cost, pentium4_cost, nocona_cost,
core2_cost, generic64_cost, generic32_cost): Set to default
untuned costs.
(k8_cost, amdfam10_cost): Costs for vectorization tuned.
(x86_builtin_vectorization_cost): New.
2007-09-10 Harsha Jagasia <harsha.jagasia@amd.com>
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c:
Change dg-final to expect 1 non-profitable loop and
3 profitable loops.
* gcc.dg/vect/costmodel/x86-64/costmodel-vect-31.c:
Change dg-final to expect 1 non-profitable loop and
3 profitable loops.
* gcc.dg/vect/costmodel/x86-64/costmodel-fast-math-vect-pr29925.c:
Change dg-final to expect 1 profitable loop.
* gcc.dg/vect/costmodel/i386/costmodel-fast-math-vect-pr29925.c:
Change dg-final to expect 1 profitable loop.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@128353 138bc75d-0d04-0410-961f-82ee72b054a4
|
|
|
2007-06-08 Harsha Jagasia <harsha.jagasia@amd.com>
Tony Linthicum <tony.linthicum@amd.com>
* doc/extend.texi: Add fvect-cost-model flag.
* common.opt (fvect-cost-model): New flag.
* tree-vectorizer.c (new_stmt_vec_info): Initialize inside and outside
cost fields in stmt_vec_info struct for STMT.
* tree-vectorizer.h (stmt_vec_info): Define inside and outside cost
fields in stmt_vec_info struct and access functions for the same.
(TARG_COND_BRANCH_COST): Define cost of conditional branch.
(TARG_VEC_STMT_COST): Define cost of any vector operation, excluding
load, store and vector to scalar operation.
(TARG_VEC_TO_SCALAR_COST): Define cost of vector to scalar operation.
(TARG_VEC_LOAD_COST): Define cost of aligned vector load.
(TARG_VEC_UNALIGNED_LOAD_COST): Define cost of misasligned vector load.
(TARG_VEC_STORE_COST): Define cost of vector store.
(vect_estimate_min_profitable_iters): Define new function.
* tree-vect-analyze.c (vect_analyze_operations): Add a compile-time
check to evaluate if loop iterations are less than minimum profitable
iterations determined by cost model or minimum vect loop bound defined
by user, whichever is more conservative.
* tree-vect-transform.c (vect_do_peeling_for_loop_bound): Add a
run-time check to evaluate if loop iterations are less than minimum
profitable iterations determined by cost model or minimum vect loop
bound defined by user, whichever is more conservative.
(vect_estimate_min_profitable_iterations): New function to estimate
mimimimum iterartions required for vector version of loop to be
profitable over scalar version.
(vect_model_reduction_cost): New function.
(vect_model_induction_cost): New function.
(vect_model_simple_cost): New function.
(vect_cost_strided_group_size): New function.
(vect_model_store_cost): New function.
(vect_model_load_cost): New function.
(vectorizable_reduction): Call vect_model_reduction_cost during
analysis phase.
(vectorizable_induction): Call vect_model_induction_cost during
analysis phase.
(vectorizable_load): Call vect_model_load_cost during analysis phase.
(vectorizable_store): Call vect_model_store_cost during analysis phase.
(vectorizable_call, vectorizable_assignment, vectorizable_operation,
vectorizable_promotion, vectorizable_demotion): Call
vect_model_simple_cost during analysis phase.
gcc/testsuite/ChangeLog:
2007-06-08 Harsha Jagasia <harsha.jagasia@amd.com>
* gcc.dg/vect/costmodel: New directory.
* gcc.dg/vect/costmodel/i386: New directory.
* gcc.dg/vect/costmodel/i386/i386-costmodel-vect.exp: New testsuite.
* gcc.dg/vect/costmodel/i386/costmodel-fast-math-vect-pr29925.c:
New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-33.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-68.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-reduc-1char.c: New test.
* gcc.dg/vect/costmodel/x86_64: New directory.
* gcc.dg/vect/costmodel/x86_64/x86_64-costmodel-vect.exp:
New testsuite.
* gcc.dg/vect/costmodel/x86_64/costmodel-fast-math-vect-pr29925.c:
New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-31.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-33.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-68.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-reduc-1char.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-pr30843.c: New test.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@125575 138bc75d-0d04-0410-961f-82ee72b054a4
|
