generic retuning part 1 - x86-tune-costs update

Richard Biener richard.guenther@gmail.com
Thu Nov 30 11:03:00 GMT 2017


On Thu, Nov 30, 2017 at 10:40 AM, Jan Hubicka <hubicka@ucw.cz> wrote:
> Hi,
> this patch makes costs in generic to math better modern chips (core, haswell,
> buldozer and zen).  The only important change is to drop cost of unaligned loads
> and stores becuase all modern chips handle it well.  This makes vectorizer to
> not peel for alignment and saves a lot of code size without sacrifying
> performance.
>
> I have benchmarked it on zen and skylake and it is small but almost consistent
> win in performance too.  Notable regression is fma3d regressing aprox. 5%
> on zen.  This is the case of native tuning as well, so I will look into it
> incrementally.
>
> Bootstrapped/regtested x86_64-linux, comitted.

The question is how we cost such things as store bandwith where IIRC
an unaligned store counts 'two' entries in the pipelines store buffers.
Likewise unaligned loads do usually still have a penalty.

What changed is that when the loads/stores happen to be aligned
using the unaligned instruction variant doesn't have a penalty.

So I'm not sure peeling for alignmend isn't a win, it just depends more
on the number of memory streams involved.

Richard.

> Honza
>
>         PR target/81616
>         * x86-tnue-costs.h (generic_cost): Revise for modern CPUs
>         * gcc.target/i386/l_fma_double_1.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_double_2.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_double_3.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_double_4.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_double_5.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_double_6.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_1.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_2.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_3.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_4.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_5.c: Update count of fma instructions.
>         * gcc.target/i386/l_fma_float_6.c: Update count of fma instructions.
> Index: config/i386/x86-tune-costs.h
> ===================================================================
> --- config/i386/x86-tune-costs.h        (revision 255252)
> +++ config/i386/x86-tune-costs.h        (working copy)
> @@ -2243,11 +2243,11 @@ struct processor_costs generic_cost = {
>     COSTS_N_INSNS (4),                  /*                               HI */
>     COSTS_N_INSNS (3),                  /*                               SI */
>     COSTS_N_INSNS (4),                  /*                               DI */
> -   COSTS_N_INSNS (2)},                 /*                            other */
> +   COSTS_N_INSNS (4)},                 /*                            other */
>    0,                                   /* cost of multiply per each bit set */
> -  {COSTS_N_INSNS (18),                 /* cost of a divide/mod for QI */
> -   COSTS_N_INSNS (26),                 /*                          HI */
> -   COSTS_N_INSNS (42),                 /*                          SI */
> +  {COSTS_N_INSNS (16),                 /* cost of a divide/mod for QI */
> +   COSTS_N_INSNS (22),                 /*                          HI */
> +   COSTS_N_INSNS (30),                 /*                          SI */
>     COSTS_N_INSNS (74),                 /*                          DI */
>     COSTS_N_INSNS (74)},                        /*                          other */
>    COSTS_N_INSNS (1),                   /* cost of movsx */
> @@ -2275,13 +2275,13 @@ struct processor_costs generic_cost = {
>    2, 3, 4,                             /* cost of moving XMM,YMM,ZMM register */
>    {6, 6, 6, 10, 15},                   /* cost of loading SSE registers
>                                            in 32,64,128,256 and 512-bit */
> -  {10, 10, 10, 15, 20},                        /* cost of unaligned loads.  */
> +  {6, 6, 6, 10, 15},                   /* cost of unaligned loads.  */
>    {6, 6, 6, 10, 15},                   /* cost of storing SSE registers
>                                            in 32,64,128,256 and 512-bit */
> -  {10, 10, 10, 15, 20},                        /* cost of unaligned storess.  */
> -  20, 20,                              /* SSE->integer and integer->SSE moves */
> -  6, 6,                                        /* Gather load static, per_elt.  */
> -  6, 6,                                        /* Gather store static, per_elt.  */
> +  {6, 6, 6, 10, 15},                   /* cost of unaligned storess.  */
> +  6, 6,                                        /* SSE->integer and integer->SSE moves */
> +  18, 6,                               /* Gather load static, per_elt.  */
> +  18, 6,                               /* Gather store static, per_elt.  */
>    32,                                  /* size of l1 cache.  */
>    512,                                 /* size of l2 cache.  */
>    64,                                  /* size of prefetch block */
> @@ -2290,11 +2290,11 @@ struct processor_costs generic_cost = {
>       value is increased to perhaps more appropriate value of 5.  */
>    3,                                   /* Branch cost */
>    COSTS_N_INSNS (3),                   /* cost of FADD and FSUB insns.  */
> -  COSTS_N_INSNS (3),                   /* cost of FMUL instruction.  */
> +  COSTS_N_INSNS (5),                   /* cost of FMUL instruction.  */
>    COSTS_N_INSNS (20),                  /* cost of FDIV instruction.  */
>    COSTS_N_INSNS (1),                   /* cost of FABS instruction.  */
>    COSTS_N_INSNS (1),                   /* cost of FCHS instruction.  */
> -  COSTS_N_INSNS (40),                  /* cost of FSQRT instruction.  */
> +  COSTS_N_INSNS (20),                  /* cost of FSQRT instruction.  */
>
>    COSTS_N_INSNS (1),                   /* cost of cheap SSE instruction.  */
>    COSTS_N_INSNS (3),                   /* cost of ADDSS/SD SUBSS/SD insns.  */
> @@ -2306,7 +2306,7 @@ struct processor_costs generic_cost = {
>    COSTS_N_INSNS (32),                  /* cost of DIVSD instruction.  */
>    COSTS_N_INSNS (30),                  /* cost of SQRTSS instruction.  */
>    COSTS_N_INSNS (58),                  /* cost of SQRTSD instruction.  */
> -  1, 2, 1, 1,                          /* reassoc int, fp, vec_int, vec_fp.  */
> +  1, 4, 3, 3,                          /* reassoc int, fp, vec_int, vec_fp.  */
>    generic_memcpy,
>    generic_memset,
>    COSTS_N_INSNS (3),                   /* cond_taken_branch_cost.  */
> Index: testsuite/gcc.target/i386/l_fma_double_1.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_1.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_1.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_double_2.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_2.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_2.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_double_3.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_3.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_3.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_double_4.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_4.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_4.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_double_5.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_5.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_5.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_double_6.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_double_6.c  (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_double_6.c  (working copy)
> @@ -13,7 +13,7 @@ typedef double adouble __attribute__((al
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+pd" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+pd" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 56 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 56 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+sd" 32 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+sd" 32 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_1.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_1.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_1.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_2.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_2.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_2.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_3.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_3.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_3.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_4.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_4.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_4.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_5.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_5.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_5.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */
> Index: testsuite/gcc.target/i386/l_fma_float_6.c
> ===================================================================
> --- testsuite/gcc.target/i386/l_fma_float_6.c   (revision 255252)
> +++ testsuite/gcc.target/i386/l_fma_float_6.c   (working copy)
> @@ -12,7 +12,7 @@
>  /* { dg-final { scan-assembler-times "vfmsub\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmadd\[123\]+ps" 8 } } */
>  /* { dg-final { scan-assembler-times "vfnmsub\[123\]+ps" 8 } } */
> -/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 120 } } */
> -/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 120 } } */
> +/* { dg-final { scan-assembler-times "vfmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfmsub\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmadd\[123\]+ss" 64 } } */
> +/* { dg-final { scan-assembler-times "vfnmsub\[123\]+ss" 64 } } */



More information about the Gcc-patches mailing list