On Tue, Aug 3, 2021 at 6:20 PM Richard Biener <richard.guent...@gmail.com> wrote: > > On Tue, Aug 3, 2021 at 11:20 AM Richard Biener > <richard.guent...@gmail.com> wrote: > > > > On Wed, Jul 28, 2021 at 4:51 AM Hongtao Liu via Gcc-patches > > <gcc-patches@gcc.gnu.org> wrote: > > > > > > On Tue, Jul 27, 2021 at 9:54 AM Hongtao Liu <crazy...@gmail.com> wrote: > > > > > > > > On Mon, Jul 26, 2021 at 4:49 PM Hongtao Liu <crazy...@gmail.com> wrote: > > > > > > > > > > Correct mail list, please reply under this email. > > > > > > > > > > On Mon, Jul 26, 2021 at 4:47 PM liuhongt <hongtao....@intel.com> > > > > > wrote: > > > > > > > > > > > > Hi: > > > > > > As decribled in PR, the pinsr instruction has poor throughput in > > > > > > SKX > > > > > > and CLX, which leads to worse performance in vectorization in some > > > > > > cases. > > > > > > This patch adds a cost member named integer_to_sse to simulate > > > > > > pinsr/movd > > > > > > which is used by vector construction, the cost is same as sse_op on > > > > > > other > > > > > > targets, but twice much as sse_op on CLX/SKX. > > > > > > Bootstrapped and regtested on x86_64-linux-gnu{-m32,}. > > > > > > Ok for trunk? > > > > > > > > > > I'm going to check in this patch if there's no objection. > > > Pushed to trunk. > > > > /* N element inserts into SSE vectors. */ > > - int cost = TYPE_VECTOR_SUBPARTS (vectype) * ix86_cost->sse_op; > > + int cost > > + = TYPE_VECTOR_SUBPARTS (vectype) * (fp ? > > + ix86_cost->sse_op > > + : > > ix86_cost->integer_to_sse); > > + > > > > so that's costing movd and pinsr the same, shouldn't we try to separate this > > by doing > > > > /* N element inserts into SSE vectors. */ > > int cost = TYPE_VECTOR_SUBPARTS (vectype) * ix86_cost->sse_op; > > /* Account for int->SSE reg moves. */ > > if (!fp) > > cost += TYPE_VECTOR_SUBPARTS (vectype) * > > ix86_cost->integer_to_sse; > > > > ? pinsr is only supported with SSE4+ IIRC. Note we also have we have pinsrw under sse2, and pinsrb/d/q w/ sse4+. integer_to_see is an estimate to model the average overhead of each integer from gpr to sse, it can be movd + unpck,or movd + pinsr. It seems reasonable to have uniform costs for scalar_to_vec and integer_to_sse. vec_to_scalar and sse_to_integer seems to be different, sse_to_integer corresponds to movd. vec_to_scalar is vec_extract. Maybe we should rename integer_to_sse to vec_set_integer. > > > > case vec_to_scalar: > > case scalar_to_vec: > > return ix86_vec_cost (mode, ix86_cost->sse_op); > > > > where scalar_to_vec is used to cost splats and vec_to_scalar is used > > to cost element extracts. Both lack costing of the move part. > > > > I realize we have GPR to XMM inserts which cover both the "move" and > > the insert but then calling this 'integer_to_sse' is a bit odd. The extract > > cost also depends on the element number for AVX2/AVX512F. The > > vectorizer usually decomposes a vector fully and never does single > > element extracts so the vextract128 cost amortizes. > > > > That said, the change leaves all targets besides skylake_cost with > > not so great defaults I think. For skylake you effectively add another > > sse_op for the int->SSE move plus '1' (for whatever reason). I think > > that's reasonable for all targets. > > > > It does look a bit odd to have > > > > 8, /* cost of moving SSE register to > > intege > > r. */ > > COSTS_N_INSNS (1), /* cost of moving integer to sse > > registe > > r. */ > > > > where sse_to_integer is used by the STV pass which mixes > > CONST_N_INSNS scaled costs and unscaled costs (ick). > > Debugging some eventually related thing I applied the same costing > as you did to skylake_cost to znver2_cost and figured that > 538.imagick_r regresses by 23% at -Ofast -march=znver2 by such > change. So it seems changes here indeed need careful benchmarking. > > Richard. > > > Richard. > > > > > > > > gcc/ChangeLog: > > > > > > > > > > > > PR target/99881 > > > > > > * config/i386/i386.h (processor_costs): Add new member > > > > > > integer_to_sse. > > > > > > * config/i386/x86-tune-costs.h (ix86_size_cost, i386_cost, > > > > > > i486_cost, pentium_cost, lakemont_cost, pentiumpro_cost, > > > > > > geode_cost, k6_cost, athlon_cost, k8_cost, amdfam10_cost, > > > > > > bdver_cost, znver1_cost, znver2_cost, znver3_cost, > > > > > > btver1_cost, btver2_cost, btver3_cost, pentium4_cost, > > > > > > nocona_cost, atom_cost, atom_cost, slm_cost, intel_cost, > > > > > > generic_cost, core_cost): Initialize integer_to_sse same > > > > > > value > > > > > > as sse_op. > > > > > > (skylake_cost): Initialize integer_to_sse twice as much as > > > > > > sse_op. > > > > > > * config/i386/i386.c (ix86_builtin_vectorization_cost): > > > > > > Use integer_to_sse instead of sse_op to calculate the cost > > > > > > of > > > > > > vec_construct. > > > > > > > > > > > > gcc/testsuite/ChangeLog: > > > > > > > > > > > > PR target/99881 > > > > > > * gcc.target/i386/pr99881.c: New test. > > > > > > --- > > > > > > gcc/config/i386/i386.c | 6 ++- > > > > > > gcc/config/i386/i386.h | 1 + > > > > > > gcc/config/i386/x86-tune-costs.h | 26 +++++++++++++ > > > > > > gcc/testsuite/gcc.target/i386/pr99881.c | 49 > > > > > > +++++++++++++++++++++++++ > > > > > > 4 files changed, 81 insertions(+), 1 deletion(-) > > > > > > create mode 100644 gcc/testsuite/gcc.target/i386/pr99881.c > > > > > > > > > > > > diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c > > > > > > index ff96134fb37..fbebd2d8f9a 100644 > > > > > > --- a/gcc/config/i386/i386.c > > > > > > +++ b/gcc/config/i386/i386.c > > > > > > @@ -22051,7 +22051,11 @@ ix86_builtin_vectorization_cost (enum > > > > > > vect_cost_for_stmt type_of_cost, > > > > > > case vec_construct: > > > > > > { > > > > > > /* N element inserts into SSE vectors. */ > > > > > > - int cost = TYPE_VECTOR_SUBPARTS (vectype) * > > > > > > ix86_cost->sse_op; > > > > > > + int cost > > > > > > + = TYPE_VECTOR_SUBPARTS (vectype) * (fp ? > > > > > > + ix86_cost->sse_op > > > > > > + : > > > > > > ix86_cost->integer_to_sse); > > > > > > + > > > > > > /* One vinserti128 for combining two SSE vectors for > > > > > > AVX256. */ > > > > > > if (GET_MODE_BITSIZE (mode) == 256) > > > > > > cost += ix86_vec_cost (mode, ix86_cost->addss); > > > > > > diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h > > > > > > index 0c2c93daf32..d1e1c225990 100644 > > > > > > --- a/gcc/config/i386/i386.h > > > > > > +++ b/gcc/config/i386/i386.h > > > > > > @@ -165,6 +165,7 @@ struct processor_costs { > > > > > > const int xmm_move, ymm_move, /* cost of moving XMM and YMM > > > > > > register. */ > > > > > > zmm_move; > > > > > > const int sse_to_integer; /* cost of moving SSE register to > > > > > > integer. */ > > > > > > + const int integer_to_sse; /* cost of moving integer to SSE > > > > > > register. */ > > > > > > const int gather_static, gather_per_elt; /* Cost of gather load > > > > > > is computed > > > > > > as static + per_item * nelts. */ > > > > > > const int scatter_static, scatter_per_elt; /* Cost of gather > > > > > > store is > > > > > > diff --git a/gcc/config/i386/x86-tune-costs.h > > > > > > b/gcc/config/i386/x86-tune-costs.h > > > > > > index ffe810f2bcb..67cfa006196 100644 > > > > > > --- a/gcc/config/i386/x86-tune-costs.h > > > > > > +++ b/gcc/config/i386/x86-tune-costs.h > > > > > > @@ -102,6 +102,7 @@ struct processor_costs ix86_size_cost = {/* > > > > > > costs for tuning for size */ > > > > > > in 128bit, 256bit and > > > > > > 512bit */ > > > > > > 3, 3, 3, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_BYTES (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 5, 0, /* Gather load > > > > > > static, per_elt. */ > > > > > > 5, 0, /* Gather store > > > > > > static, per_elt. */ > > > > > > 0, /* size of l1 cache */ > > > > > > @@ -211,6 +212,7 @@ struct processor_costs i386_cost = { /* > > > > > > 386 specific costs */ > > > > > > {4, 8, 16, 32, 64}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 0, /* size of l1 cache */ > > > > > > @@ -319,6 +321,7 @@ struct processor_costs i486_cost = { /* > > > > > > 486 specific costs */ > > > > > > {4, 8, 16, 32, 64}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 4, /* size of l1 cache. 486 > > > > > > has 8kB cache > > > > > > @@ -429,6 +432,7 @@ struct processor_costs pentium_cost = { > > > > > > {4, 8, 16, 32, 64}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 8, /* size of l1 cache. */ > > > > > > @@ -530,6 +534,7 @@ struct processor_costs lakemont_cost = { > > > > > > {4, 8, 16, 32, 64}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 8, /* size of l1 cache. */ > > > > > > @@ -646,6 +651,7 @@ struct processor_costs pentiumpro_cost = { > > > > > > {4, 8, 16, 32, 64}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 8, /* size of l1 cache. */ > > > > > > @@ -753,6 +759,7 @@ struct processor_costs geode_cost = { > > > > > > {2, 2, 8, 16, 32}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 2, 2, /* Gather load > > > > > > static, per_elt. */ > > > > > > 2, 2, /* Gather store > > > > > > static, per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -860,6 +867,7 @@ struct processor_costs k6_cost = { > > > > > > {2, 2, 8, 16, 32}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 2, 2, /* Gather load > > > > > > static, per_elt. */ > > > > > > 2, 2, /* Gather store > > > > > > static, per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -973,6 +981,7 @@ struct processor_costs athlon_cost = { > > > > > > {4, 4, 10, 10, 20}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 5, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -1088,6 +1097,7 @@ struct processor_costs k8_cost = { > > > > > > {4, 4, 10, 10, 20}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 5, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -1216,6 +1226,7 @@ struct processor_costs amdfam10_cost = { > > > > > > {4, 4, 5, 10, 20}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 3, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 4, 4, /* Gather load > > > > > > static, per_elt. */ > > > > > > 4, 4, /* Gather store > > > > > > static, per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -1336,6 +1347,7 @@ const struct processor_costs bdver_cost = { > > > > > > {10, 10, 10, 40, 60}, /* cost of > > > > > > unaligned stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 16, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 12, 12, /* Gather load static, > > > > > > per_elt. */ > > > > > > 10, 10, /* Gather store static, > > > > > > per_elt. */ > > > > > > 16, /* size of l1 cache. */ > > > > > > @@ -1477,6 +1489,7 @@ struct processor_costs znver1_cost = { > > > > > > {8, 8, 8, 16, 32}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 3, 6, /* cost of moving > > > > > > XMM,YMM,ZMM register. */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > /* VGATHERDPD is 23 uops and throughput is 9, VGATHERDPD is 35 > > > > > > uops, > > > > > > throughput 12. Approx 9 uops do not depend on vector size > > > > > > and every load > > > > > > is 7 uops. */ > > > > > > @@ -1633,6 +1646,7 @@ struct processor_costs znver2_cost = { > > > > > > 2, 2, 3, /* cost of moving > > > > > > XMM,YMM,ZMM > > > > > > register. */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > /* VGATHERDPD is 23 uops and throughput is 9, VGATHERDPD is 35 > > > > > > uops, > > > > > > throughput 12. Approx 9 uops do not depend on vector size > > > > > > and every load > > > > > > is 7 uops. */ > > > > > > @@ -1765,6 +1779,7 @@ struct processor_costs znver3_cost = { > > > > > > 2, 2, 3, /* cost of moving > > > > > > XMM,YMM,ZMM > > > > > > register. */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > /* VGATHERDPD is 15 uops and throughput is 4, VGATHERDPS is 23 > > > > > > uops, > > > > > > throughput 9. Approx 7 uops do not depend on vector size and > > > > > > every load > > > > > > is 4 uops. */ > > > > > > @@ -1909,6 +1924,7 @@ struct processor_costs skylake_cost = { > > > > > > {8, 8, 8, 8, 16}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 2, 4, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2)+1, /* cost of moving integer > > > > > > to sse register. */ > > > > > > 20, 8, /* Gather load static, > > > > > > per_elt. */ > > > > > > 22, 10, /* Gather store static, > > > > > > per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -2035,6 +2051,7 @@ struct processor_costs icelake_cost = { > > > > > > {8, 8, 8, 8, 16}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 2, 4, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 20, 8, /* Gather load static, > > > > > > per_elt. */ > > > > > > 22, 10, /* Gather store static, > > > > > > per_elt. */ > > > > > > 64, /* size of l1 cache. */ > > > > > > @@ -2148,6 +2165,7 @@ const struct processor_costs btver1_cost = { > > > > > > {10, 10, 12, 48, 96}, /* cost of > > > > > > unaligned stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 14, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 10, 10, /* Gather load static, > > > > > > per_elt. */ > > > > > > 10, 10, /* Gather store static, > > > > > > per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -2258,6 +2276,7 @@ const struct processor_costs btver2_cost = { > > > > > > {10, 10, 12, 48, 96}, /* cost of > > > > > > unaligned stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 14, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 10, 10, /* Gather load static, > > > > > > per_elt. */ > > > > > > 10, 10, /* Gather store static, > > > > > > per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -2367,6 +2386,7 @@ struct processor_costs pentium4_cost = { > > > > > > {32, 32, 32, 64, 128}, /* cost of unaligned > > > > > > stores. */ > > > > > > 12, 24, 48, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 20, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 16, 16, /* Gather load static, > > > > > > per_elt. */ > > > > > > 16, 16, /* Gather store static, > > > > > > per_elt. */ > > > > > > 8, /* size of l1 cache. */ > > > > > > @@ -2479,6 +2499,7 @@ struct processor_costs nocona_cost = { > > > > > > {24, 24, 24, 48, 96}, /* cost of > > > > > > unaligned stores. */ > > > > > > 6, 12, 24, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 20, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (2), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 12, 12, /* Gather load static, > > > > > > per_elt. */ > > > > > > 12, 12, /* Gather store static, > > > > > > per_elt. */ > > > > > > 8, /* size of l1 cache. */ > > > > > > @@ -2589,6 +2610,7 @@ struct processor_costs atom_cost = { > > > > > > {16, 16, 16, 32, 64}, /* cost of > > > > > > unaligned stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 8, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 8, 8, /* Gather load > > > > > > static, per_elt. */ > > > > > > 8, 8, /* Gather store > > > > > > static, per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -2699,6 +2721,7 @@ struct processor_costs slm_cost = { > > > > > > {16, 16, 16, 32, 64}, /* cost of > > > > > > unaligned stores. */ > > > > > > 2, 4, 8, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 8, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 8, 8, /* Gather load > > > > > > static, per_elt. */ > > > > > > 8, 8, /* Gather store > > > > > > static, per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -2809,6 +2832,7 @@ struct processor_costs intel_cost = { > > > > > > {10, 10, 10, 10, 10}, /* cost of > > > > > > unaligned loads. */ > > > > > > 2, 2, 2, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 4, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 6, 6, /* Gather load > > > > > > static, per_elt. */ > > > > > > 6, 6, /* Gather store > > > > > > static, per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -2926,6 +2950,7 @@ struct processor_costs generic_cost = { > > > > > > {6, 6, 6, 10, 15}, /* cost of unaligned > > > > > > storess. */ > > > > > > 2, 3, 4, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 6, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > 18, 6, /* Gather load static, > > > > > > per_elt. */ > > > > > > 18, 6, /* Gather store static, > > > > > > per_elt. */ > > > > > > 32, /* size of l1 cache. */ > > > > > > @@ -3049,6 +3074,7 @@ struct processor_costs core_cost = { > > > > > > {6, 6, 6, 6, 12}, /* cost of unaligned > > > > > > stores. */ > > > > > > 2, 2, 4, /* cost of moving > > > > > > XMM,YMM,ZMM register */ > > > > > > 2, /* cost of moving SSE > > > > > > register to integer. */ > > > > > > + COSTS_N_INSNS (1), /* cost of moving integer > > > > > > to sse register. */ > > > > > > /* VGATHERDPD is 7 uops, rec throughput 5, while VGATHERDPD is 9 > > > > > > uops, > > > > > > rec. throughput 6. > > > > > > So 5 uops statically and one uops per load. */ > > > > > > diff --git a/gcc/testsuite/gcc.target/i386/pr99881.c > > > > > > b/gcc/testsuite/gcc.target/i386/pr99881.c > > > > > > new file mode 100644 > > > > > > index 00000000000..7ae51c8310d > > > > > > --- /dev/null > > > > > > +++ b/gcc/testsuite/gcc.target/i386/pr99881.c > > > > > > @@ -0,0 +1,49 @@ > > > > > > +/* PR target/99881. */ > > > > > > +/* { dg-do compile } */ > > > > > > +/* { dg-options "-Ofast -march=skylake" } */ > > > > > > +/* { dg-final { scan-assembler-not "xmm[0-9]" } } */ > > > > > > + > > > > > > +void > > > > > > +foo (int* __restrict a, int n, int c) > > > > > > +{ > > > > > > + a[0] = n; > > > > > > + a[1] = c; > > > > > > +} > > > > > > + > > > > > > +void > > > > > > +foo1 (int* __restrict a, int n, int b, int c, int d) > > > > > > +{ > > > > > > + a[0] = n; > > > > > > + a[1] = b; > > > > > > + a[2] = c; > > > > > > + a[3] = d; > > > > > > +} > > > > > > + > > > > > > +void > > > > > > +foo2 (int* __restrict a, int n, int b, int c, int d, int e, int f, > > > > > > int g, int h) > > > > > > +{ > > > > > > + a[0] = n; > > > > > > + a[1] = b; > > > > > > + a[2] = c; > > > > > > + a[3] = d; > > > > > > + a[4] = e; > > > > > > + a[5] = f; > > > > > > + a[6] = g; > > > > > > + a[7] = h; > > > > > > +} > > > > > > + > > > > > > +void > > > > > > +foo3 (long long* __restrict a, long long n, long long c) > > > > > > +{ > > > > > > + a[0] = n; > > > > > > + a[1] = c; > > > > > > +} > > > > > > + > > > > > > +void > > > > > > +foo4 (long long* __restrict a, long long n, long long b, long long > > > > > > c, long long d) > > > > > > +{ > > > > > > + a[0] = n; > > > > > > + a[1] = b; > > > > > > + a[2] = c; > > > > > > + a[3] = d; > > > > > > +} > > > > > > -- > > > > > > 2.18.1 > > > > > > > > > > > > > > > > > > > > > -- > > > > > BR, > > > > > Hongtao > > > > > > > > > > > > > > > > -- > > > > BR, > > > > Hongtao > > > > > > > > > > > > -- > > > BR, > > > Hongtao
-- BR, Hongtao