This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Zen tuning part 10: ix86_builtin_vectorization_cost fixes
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: gcc-patches at gcc dot gnu dot org, Venkataramanan dot Kumar at amd dot com, ubizjak at gmail dot com, hongjiu dot lu at intel dot com, rguenther at suse dot de
- Date: Mon, 23 Oct 2017 10:36:28 +0200
- Subject: Zen tuning part 10: ix86_builtin_vectorization_cost fixes
- Authentication-results: sourceware.org; auth=none
Hi,
this is patch to ix86_builtin_vectorization_cost I have comitted. Compared to
earlier version it only removes now unused fields in processor_costs. The
patch improves facerec on all tested targets (amdfam10, Buldozer, Zen and
core), largest improvement is on Zen by about 25% (for core improvement is 5%).
It also improves gromacs and lbm (on Zen and core) and makes small regression
in gamess (sub 1%) and tonto (5-7%). Tonto is fixed by the scatter,gather patch
I plan as a followup.
There are also improvements for poyhedron (fatigue, fatigue2 6%, test_fpu 3%)
There is a hack of making unaligned store twice the cost of aligned to make it
similar to previous costmodel. With this and the earlier fixes to move cost
tables it no longer causes regressions in testsuite with exception of
gcc.target/i386/pr79683.c where costmodel now claims that vectorization is not
profitable with generic (it is profitable i.e. for core) which seems corect:
struct s {
__INT64_TYPE__ a;
__INT64_TYPE__ b;
};
void test(struct s __seg_gs *x) {
x->a += 1;
x->b -= 1;
}
We model vector integer ops as more expensive then integer operations.
I disabled costmodel there.
The unaligned and avx costs will be fixed as followup.
Bootstrapped/regtested x86_64-linux, comitted.
Honza
* gcc.target/i386/pr79683.c: Disable costmodel.
* i386.c (ix86_builtin_vectorization_cost): Use existing rtx_cost
latencies instead of having separate table; make difference between
integer and float costs.
* i386.h (processor_costs): Remove scalar_stmt_cost,
scalar_load_cost, scalar_store_cost, vec_stmt_cost, vec_to_scalar_cost,
scalar_to_vec_cost, vec_align_load_cost, vec_unalign_load_cost,
vec_store_cost.
* x86-tune-costs.h: Remove entries which has been removed in
procesor_costs from all tables; make cond_taken_branch_cost
and cond_not_taken_branch_cost COST_N_INSNS based.
Index: testsuite/gcc.target/i386/pr79683.c
===================================================================
--- testsuite/gcc.target/i386/pr79683.c (revision 253957)
+++ testsuite/gcc.target/i386/pr79683.c (working copy)
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O3 -msse2" } */
+/* { dg-options "-O3 -msse2 -fvect-cost-model=unlimited" } */
struct s {
__INT64_TYPE__ a;
Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c (revision 253957)
+++ config/i386/i386.c (working copy)
@@ -44051,37 +44051,61 @@ static int
ix86_builtin_vectorization_cost (enum vect_cost_for_stmt type_of_cost,
tree vectype, int)
{
+ bool fp = false;
+ machine_mode mode = TImode;
+ if (vectype != NULL)
+ {
+ fp = FLOAT_TYPE_P (vectype);
+ mode = TYPE_MODE (vectype);
+ }
+
switch (type_of_cost)
{
case scalar_stmt:
- return ix86_cost->scalar_stmt_cost;
+ return fp ? ix86_cost->addss : COSTS_N_INSNS (1);
case scalar_load:
- return ix86_cost->scalar_load_cost;
+ /* load/store costs are relative to register move which is 2. Recompute
+ it to COSTS_N_INSNS so everything have same base. */
+ return COSTS_N_INSNS (fp ? ix86_cost->sse_load[0]
+ : ix86_cost->int_load [2]) / 2;
case scalar_store:
- return ix86_cost->scalar_store_cost;
+ return COSTS_N_INSNS (fp ? ix86_cost->sse_store[0]
+ : ix86_cost->int_store [2]) / 2;
case vector_stmt:
- return ix86_cost->vec_stmt_cost;
+ return ix86_vec_cost (mode,
+ fp ? ix86_cost->addss : ix86_cost->sse_op,
+ true);
case vector_load:
- return ix86_cost->vec_align_load_cost;
+ return ix86_vec_cost (mode,
+ COSTS_N_INSNS (ix86_cost->sse_load[2]) / 2,
+ true);
case vector_store:
- return ix86_cost->vec_store_cost;
+ return ix86_vec_cost (mode,
+ COSTS_N_INSNS (ix86_cost->sse_store[2]) / 2,
+ true);
case vec_to_scalar:
- return ix86_cost->vec_to_scalar_cost;
-
case scalar_to_vec:
- return ix86_cost->scalar_to_vec_cost;
+ return ix86_vec_cost (mode, ix86_cost->sse_op, true);
+ /* We should have separate costs for unaligned loads and gather/scatter.
+ Do that incrementally. */
case unaligned_load:
- case unaligned_store:
case vector_gather_load:
+ return ix86_vec_cost (mode,
+ COSTS_N_INSNS (ix86_cost->sse_load[2]),
+ true);
+
+ case unaligned_store:
case vector_scatter_store:
- return ix86_cost->vec_unalign_load_cost;
+ return ix86_vec_cost (mode,
+ COSTS_N_INSNS (ix86_cost->sse_store[2]),
+ true);
case cond_branch_taken:
return ix86_cost->cond_taken_branch_cost;
@@ -44091,10 +44115,11 @@ ix86_builtin_vectorization_cost (enum ve
case vec_perm:
case vec_promote_demote:
- return ix86_cost->vec_stmt_cost;
+ return ix86_vec_cost (mode,
+ ix86_cost->sse_op, true);
case vec_construct:
- return ix86_cost->vec_stmt_cost * (TYPE_VECTOR_SUBPARTS (vectype) - 1);
+ return ix86_vec_cost (mode, ix86_cost->sse_op, false);
default:
gcc_unreachable ();
Index: config/i386/i386.h
===================================================================
--- config/i386/i386.h (revision 253957)
+++ config/i386/i386.h (working copy)
@@ -277,18 +277,6 @@ struct processor_costs {
parallel. See also
ix86_reassociation_width. */
struct stringop_algs *memcpy, *memset;
- const int scalar_stmt_cost; /* Cost of any scalar operation, excluding
- load and store. */
- const int scalar_load_cost; /* Cost of scalar load. */
- const int scalar_store_cost; /* Cost of scalar store. */
- const int vec_stmt_cost; /* Cost of any vector operation, excluding
- load, store, vector-to-scalar and
- scalar-to-vector operation. */
- const int vec_to_scalar_cost; /* Cost of vect-to-scalar operation. */
- const int scalar_to_vec_cost; /* Cost of scalar-to-vector operation. */
- const int vec_align_load_cost; /* Cost of aligned vector load. */
- const int vec_unalign_load_cost; /* Cost of unaligned vector load. */
- const int vec_store_cost; /* Cost of vector store. */
const int cond_taken_branch_cost; /* Cost of taken branch for vectorizer
cost model. */
const int cond_not_taken_branch_cost;/* Cost of not taken branch for
Index: config/i386/x86-tune-costs.h
===================================================================
--- config/i386/x86-tune-costs.h (revision 253958)
+++ config/i386/x86-tune-costs.h (working copy)
@@ -79,17 +79,8 @@ struct processor_costs ix86_size_cost =
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
ix86_size_memcpy,
ix86_size_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 1, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 1, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_BYTES (1), /* cond_taken_branch_cost. */
+ COSTS_N_BYTES (1), /* cond_not_taken_branch_cost. */
};
/* Processor costs (relative to an add) */
@@ -167,17 +158,8 @@ struct processor_costs i386_cost = { /*
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
i386_memcpy,
i386_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs i486_memcpy[2] = {
@@ -256,17 +238,8 @@ struct processor_costs i486_cost = { /*
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
i486_memcpy,
i486_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs pentium_memcpy[2] = {
@@ -343,17 +316,8 @@ struct processor_costs pentium_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
pentium_memcpy,
pentium_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static const
@@ -423,17 +387,8 @@ struct processor_costs lakemont_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
pentium_memcpy,
pentium_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* PentiumPro has optimized rep instructions for blocks aligned by 8 bytes
@@ -518,17 +473,8 @@ struct processor_costs pentiumpro_cost =
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
pentiumpro_memcpy,
pentiumpro_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs geode_memcpy[2] = {
@@ -605,17 +551,8 @@ struct processor_costs geode_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
geode_memcpy,
geode_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs k6_memcpy[2] = {
@@ -694,17 +631,8 @@ struct processor_costs k6_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
k6_memcpy,
k6_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* For some reason, Athlon deals better with REP prefix (relative to loops)
@@ -784,17 +712,8 @@ struct processor_costs athlon_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
athlon_memcpy,
athlon_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* K8 has optimized REP instruction for medium sized blocks, but for very
@@ -883,17 +802,8 @@ struct processor_costs k8_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
k8_memcpy,
k8_memset,
- 4, /* scalar_stmt_cost. */
- 2, /* scalar load_cost. */
- 2, /* scalar_store_cost. */
- 5, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 2, /* vec_align_load_cost. */
- 3, /* vec_unalign_load_cost. */
- 3, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
/* AMDFAM10 has optimized REP instruction for medium sized blocks, but for
@@ -989,17 +899,8 @@ struct processor_costs amdfam10_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
amdfam10_memcpy,
amdfam10_memset,
- 4, /* scalar_stmt_cost. */
- 2, /* scalar load_cost. */
- 2, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 2, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 2, /* vec_store_cost. */
- 2, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* BDVER1 has optimized REP instruction for medium sized blocks, but for
@@ -1097,17 +998,8 @@ const struct processor_costs bdver1_cost
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
bdver1_memcpy,
bdver1_memset,
- 6, /* scalar_stmt_cost. */
- 4, /* scalar load_cost. */
- 4, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 4, /* vec_align_load_cost. */
- 4, /* vec_unalign_load_cost. */
- 4, /* vec_store_cost. */
- 4, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (4), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
/* BDVER2 has optimized REP instruction for medium sized blocks, but for
@@ -1206,17 +1098,8 @@ const struct processor_costs bdver2_cost
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
bdver2_memcpy,
bdver2_memset,
- 6, /* scalar_stmt_cost. */
- 4, /* scalar load_cost. */
- 4, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 4, /* vec_align_load_cost. */
- 4, /* vec_unalign_load_cost. */
- 4, /* vec_store_cost. */
- 4, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (4), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
@@ -1306,17 +1189,8 @@ struct processor_costs bdver3_cost = {
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
bdver3_memcpy,
bdver3_memset,
- 6, /* scalar_stmt_cost. */
- 4, /* scalar load_cost. */
- 4, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 4, /* vec_align_load_cost. */
- 4, /* vec_unalign_load_cost. */
- 4, /* vec_store_cost. */
- 4, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (4), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
/* BDVER4 has optimized REP instruction for medium sized blocks, but for
@@ -1405,17 +1279,8 @@ struct processor_costs bdver4_cost = {
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
bdver4_memcpy,
bdver4_memset,
- 6, /* scalar_stmt_cost. */
- 4, /* scalar load_cost. */
- 4, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 4, /* vec_align_load_cost. */
- 4, /* vec_unalign_load_cost. */
- 4, /* vec_store_cost. */
- 4, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (4), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
@@ -1524,17 +1389,8 @@ struct processor_costs znver1_cost = {
4, 4, 3, 6, /* reassoc int, fp, vec_int, vec_fp. */
znver1_memcpy,
znver1_memset,
- 6, /* scalar_stmt_cost. */
- 4, /* scalar load_cost. */
- 4, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 4, /* vec_align_load_cost. */
- 4, /* vec_unalign_load_cost. */
- 4, /* vec_store_cost. */
- 4, /* cond_taken_branch_cost. */
- 2, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (4), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_not_taken_branch_cost. */
};
/* BTVER1 has optimized REP instruction for medium sized blocks, but for
@@ -1624,17 +1480,8 @@ const struct processor_costs btver1_cost
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
btver1_memcpy,
btver1_memset,
- 4, /* scalar_stmt_cost. */
- 2, /* scalar load_cost. */
- 2, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 2, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 2, /* vec_store_cost. */
- 2, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs btver2_memcpy[2] = {
@@ -1721,17 +1568,8 @@ const struct processor_costs btver2_cost
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
btver2_memcpy,
btver2_memset,
- 4, /* scalar_stmt_cost. */
- 2, /* scalar load_cost. */
- 2, /* scalar_store_cost. */
- 6, /* vec_stmt_cost. */
- 0, /* vec_to_scalar_cost. */
- 2, /* scalar_to_vec_cost. */
- 2, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 2, /* vec_store_cost. */
- 2, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (2), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs pentium4_memcpy[2] = {
@@ -1809,17 +1647,8 @@ struct processor_costs pentium4_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
pentium4_memcpy,
pentium4_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs nocona_memcpy[2] = {
@@ -1900,17 +1729,8 @@ struct processor_costs nocona_cost = {
1, 1, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
nocona_memcpy,
nocona_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs atom_memcpy[2] = {
@@ -1989,17 +1809,8 @@ struct processor_costs atom_cost = {
2, 2, 2, 2, /* reassoc int, fp, vec_int, vec_fp. */
atom_memcpy,
atom_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs slm_memcpy[2] = {
@@ -2078,17 +1889,8 @@ struct processor_costs slm_cost = {
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
slm_memcpy,
slm_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 4, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
static stringop_algs intel_memcpy[2] = {
@@ -2167,17 +1969,8 @@ struct processor_costs intel_cost = {
1, 4, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
intel_memcpy,
intel_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 4, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* Generic should produce code tuned for Core-i7 (and newer chips)
@@ -2265,17 +2058,8 @@ struct processor_costs generic_cost = {
1, 2, 1, 1, /* reassoc int, fp, vec_int, vec_fp. */
generic_memcpy,
generic_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};
/* core_cost should produce code tuned for Core familly of CPUs. */
@@ -2366,16 +2150,7 @@ struct processor_costs core_cost = {
1, 4, 2, 2, /* reassoc int, fp, vec_int, vec_fp. */
core_memcpy,
core_memset,
- 1, /* scalar_stmt_cost. */
- 1, /* scalar load_cost. */
- 1, /* scalar_store_cost. */
- 1, /* vec_stmt_cost. */
- 1, /* vec_to_scalar_cost. */
- 1, /* scalar_to_vec_cost. */
- 1, /* vec_align_load_cost. */
- 2, /* vec_unalign_load_cost. */
- 1, /* vec_store_cost. */
- 3, /* cond_taken_branch_cost. */
- 1, /* cond_not_taken_branch_cost. */
+ COSTS_N_INSNS (3), /* cond_taken_branch_cost. */
+ COSTS_N_INSNS (1), /* cond_not_taken_branch_cost. */
};