From eb23241ba81aace0c881ccee4643632809741953 Mon Sep 17 00:00:00 2001 From: Richard Sandiford Date: Sat, 16 Nov 2019 10:43:52 +0000 Subject: [PATCH] [AArch64] Enable VECT_COMPARE_COSTS by default for SVE This patch enables VECT_COMPARE_COSTS by default for SVE, both so that we can compare SVE against Advanced SIMD and so that (with future patches) we can compare multiple SVE vectorisation approaches against each other. It also adds a target-specific --param to control this. 2019-11-16 Richard Sandiford gcc/ * config/aarch64/aarch64.opt (--param=aarch64-sve-compare-costs): New option. * doc/invoke.texi: Document it. * config/aarch64/aarch64.c (aarch64_autovectorize_vector_modes): By default, return VECT_COMPARE_COSTS for SVE. gcc/testsuite/ * gcc.target/aarch64/sve/reduc_3.c: Split multi-vector cases out into... * gcc.target/aarch64/sve/reduc_3_costly.c: ...this new test, passing -fno-vect-cost-model for them. * gcc.target/aarch64/sve/slp_6.c: Add -fno-vect-cost-model. * gcc.target/aarch64/sve/slp_7.c, * gcc.target/aarch64/sve/slp_7_run.c: Split multi-vector cases out into... * gcc.target/aarch64/sve/slp_7_costly.c, * gcc.target/aarch64/sve/slp_7_costly_run.c: ...these new tests, passing -fno-vect-cost-model for them. * gcc.target/aarch64/sve/while_7.c: Add -fno-vect-cost-model. * gcc.target/aarch64/sve/while_9.c: Likewise. From-SVN: r278337 --- gcc/ChangeLog | 8 ++++ gcc/config/aarch64/aarch64.c | 10 ++++- gcc/config/aarch64/aarch64.opt | 4 ++ gcc/doc/invoke.texi | 18 +++++++- gcc/testsuite/ChangeLog | 16 +++++++ .../gcc.target/aarch64/sve/reduc_3.c | 26 +++++------ .../gcc.target/aarch64/sve/reduc_3_costly.c | 32 ++++++++++++++ gcc/testsuite/gcc.target/aarch64/sve/slp_6.c | 2 +- gcc/testsuite/gcc.target/aarch64/sve/slp_7.c | 16 ++----- .../gcc.target/aarch64/sve/slp_7_costly.c | 43 +++++++++++++++++++ .../gcc.target/aarch64/sve/slp_7_costly_run.c | 5 +++ .../gcc.target/aarch64/sve/slp_7_run.c | 6 ++- .../gcc.target/aarch64/sve/while_7.c | 2 +- .../gcc.target/aarch64/sve/while_9.c | 2 +- 14 files changed, 154 insertions(+), 36 deletions(-) create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/reduc_3_costly.c create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly.c create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly_run.c diff --git a/gcc/ChangeLog b/gcc/ChangeLog index f809def56c4..2eee46b4ebd 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,11 @@ +2019-11-16 Richard Sandiford + + * config/aarch64/aarch64.opt (--param=aarch64-sve-compare-costs): + New option. + * doc/invoke.texi: Document it. + * config/aarch64/aarch64.c (aarch64_autovectorize_vector_modes): + By default, return VECT_COMPARE_COSTS for SVE. + 2019-11-16 Richard Sandiford * target.h (VECT_COMPARE_COSTS): New constant. diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c index e2251a2ef32..9ffe2138c63 100644 --- a/gcc/config/aarch64/aarch64.c +++ b/gcc/config/aarch64/aarch64.c @@ -15962,7 +15962,15 @@ aarch64_autovectorize_vector_modes (vector_modes *modes, bool) for this case. */ modes->safe_push (V2SImode); - return 0; + unsigned int flags = 0; + /* Consider enabling VECT_COMPARE_COSTS for SVE, both so that we + can compare SVE against Advanced SIMD and so that we can compare + multiple SVE vectorization approaches against each other. There's + not really any point doing this for Advanced SIMD only, since the + first mode that works should always be the best. */ + if (TARGET_SVE && aarch64_sve_compare_costs) + flags |= VECT_COMPARE_COSTS; + return flags; } /* Implement TARGET_MANGLE_TYPE. */ diff --git a/gcc/config/aarch64/aarch64.opt b/gcc/config/aarch64/aarch64.opt index fc43428b32a..3b675e16259 100644 --- a/gcc/config/aarch64/aarch64.opt +++ b/gcc/config/aarch64/aarch64.opt @@ -258,3 +258,7 @@ long aarch64_stack_protector_guard_offset = 0 moutline-atomics Target Report Mask(OUTLINE_ATOMICS) Save Generate local calls to out-of-line atomic operations. + +-param=aarch64-sve-compare-costs= +Target Joined UInteger Var(aarch64_sve_compare_costs) Init(1) IntegerRange(0, 1) Param +When vectorizing for SVE, consider using unpacked vectors for smaller elements and use the cost model to pick the cheapest approach. Also use the cost model to choose between SVE and Advanced SIMD vectorization. diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi index fe79ca2247a..7f19d675088 100644 --- a/gcc/doc/invoke.texi +++ b/gcc/doc/invoke.texi @@ -11179,8 +11179,8 @@ without notice in future releases. In order to get minimal, maximal and default value of a parameter, one can use @option{--help=param -Q} options. -In each case, the @var{value} is an integer. The allowable choices for -@var{name} are: +In each case, the @var{value} is an integer. The following choices +of @var{name} are recognized for all targets: @table @gcctabopt @item predictable-branch-outcome @@ -12396,6 +12396,20 @@ statements or when determining their validity prior to issuing diagnostics. @end table + +The following choices of @var{name} are available on AArch64 targets: + +@table @gcctabopt +@item aarch64-sve-compare-costs +When vectorizing for SVE, consider using ``unpacked'' vectors for +smaller elements and use the cost model to pick the cheapest approach. +Also use the cost model to choose between SVE and Advanced SIMD vectorization. + +Using unpacked vectors includes storing smaller elements in larger +containers and accessing elements with extending loads and truncating +stores. +@end table + @end table @node Instrumentation Options diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index daac2708d7f..4274edd0846 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,19 @@ +2019-11-16 Richard Sandiford + + * gcc.target/aarch64/sve/reduc_3.c: Split multi-vector cases out + into... + * gcc.target/aarch64/sve/reduc_3_costly.c: ...this new test, + passing -fno-vect-cost-model for them. + * gcc.target/aarch64/sve/slp_6.c: Add -fno-vect-cost-model. + * gcc.target/aarch64/sve/slp_7.c, + * gcc.target/aarch64/sve/slp_7_run.c: Split multi-vector cases out + into... + * gcc.target/aarch64/sve/slp_7_costly.c, + * gcc.target/aarch64/sve/slp_7_costly_run.c: ...these new tests, + passing -fno-vect-cost-model for them. + * gcc.target/aarch64/sve/while_7.c: Add -fno-vect-cost-model. + * gcc.target/aarch64/sve/while_9.c: Likewise. + 2019-11-16 Richard Sandiford * gcc.dg/vect/bb-slp-4.c: Expect the block to be vectorized diff --git a/gcc/testsuite/gcc.target/aarch64/sve/reduc_3.c b/gcc/testsuite/gcc.target/aarch64/sve/reduc_3.c index 456119978ef..0fc193b52e7 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/reduc_3.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/reduc_3.c @@ -17,7 +17,6 @@ void reduc_ptr_##DSTTYPE##_##SRCTYPE (DSTTYPE *restrict sum, \ REDUC_PTR (int8_t, int8_t) REDUC_PTR (int16_t, int16_t) - REDUC_PTR (int32_t, int32_t) REDUC_PTR (int64_t, int64_t) @@ -25,17 +24,6 @@ REDUC_PTR (_Float16, _Float16) REDUC_PTR (float, float) REDUC_PTR (double, double) -/* Widening reductions. */ -REDUC_PTR (int32_t, int8_t) -REDUC_PTR (int32_t, int16_t) - -REDUC_PTR (int64_t, int8_t) -REDUC_PTR (int64_t, int16_t) -REDUC_PTR (int64_t, int32_t) - -REDUC_PTR (float, _Float16) -REDUC_PTR (double, float) - /* Float<>Int conversions */ REDUC_PTR (_Float16, int16_t) REDUC_PTR (float, int32_t) @@ -45,8 +33,14 @@ REDUC_PTR (int16_t, _Float16) REDUC_PTR (int32_t, float) REDUC_PTR (int64_t, double) -/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s\n} 3 } } */ -/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 4 } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.h\n} 2 { xfail *-*-* } } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s\n} 2 { xfail *-*-* } } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 2 { xfail *-*-* } } } */ +/* We don't yet vectorize the int<-float cases. */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */ /* { dg-final { scan-assembler-times {\tfaddv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 2 } } */ -/* { dg-final { scan-assembler-times {\tfaddv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 3 } } */ -/* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 3 } } */ +/* { dg-final { scan-assembler-times {\tfaddv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 2 } } */ +/* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 2 } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/sve/reduc_3_costly.c b/gcc/testsuite/gcc.target/aarch64/sve/reduc_3_costly.c new file mode 100644 index 00000000000..988459df8af --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/reduc_3_costly.c @@ -0,0 +1,32 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -ftree-vectorize -ffast-math -fno-vect-cost-model" } */ + +#include + +#define NUM_ELEMS(TYPE) (32 / sizeof (TYPE)) + +#define REDUC_PTR(DSTTYPE, SRCTYPE) \ +void reduc_ptr_##DSTTYPE##_##SRCTYPE (DSTTYPE *restrict sum, \ + SRCTYPE *restrict array, \ + int count) \ +{ \ + *sum = 0; \ + for (int i = 0; i < count; ++i) \ + *sum += array[i]; \ +} + +/* Widening reductions. */ +REDUC_PTR (int32_t, int8_t) +REDUC_PTR (int32_t, int16_t) + +REDUC_PTR (int64_t, int8_t) +REDUC_PTR (int64_t, int16_t) +REDUC_PTR (int64_t, int32_t) + +REDUC_PTR (float, _Float16) +REDUC_PTR (double, float) + +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s\n} 2 } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 3 } } */ +/* { dg-final { scan-assembler-times {\tfaddv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */ +/* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/sve/slp_6.c b/gcc/testsuite/gcc.target/aarch64/sve/slp_6.c index 80fa3502f70..44d128477d2 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/slp_6.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/slp_6.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable -ffast-math" } */ +/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable -ffast-math -fno-vect-cost-model" } */ #include diff --git a/gcc/testsuite/gcc.target/aarch64/sve/slp_7.c b/gcc/testsuite/gcc.target/aarch64/sve/slp_7.c index dbc32a4e798..19207207999 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/slp_7.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/slp_7.c @@ -31,37 +31,27 @@ vec_slp_##TYPE (TYPE *restrict a, TYPE *restrict b, int n) \ T (uint16_t) \ T (int32_t) \ T (uint32_t) \ - T (int64_t) \ - T (uint64_t) \ T (_Float16) \ - T (float) \ - T (double) + T (float) TEST_ALL (VEC_PERM) -/* We can't use SLP for the 64-bit loops, since the number of reduction - results might be greater than the number of elements in the vector. - Otherwise we have two loads per loop, one for the initial vector - and one for the loop body. */ +/* We have two loads per loop, one for the initial vector and one for + the loop body. */ /* { dg-final { scan-assembler-times {\tld1b\t} 2 } } */ /* { dg-final { scan-assembler-times {\tld1h\t} 3 } } */ /* { dg-final { scan-assembler-times {\tld1w\t} 3 } } */ -/* { dg-final { scan-assembler-times {\tld4d\t} 3 } } */ /* { dg-final { scan-assembler-not {\tld4b\t} } } */ /* { dg-final { scan-assembler-not {\tld4h\t} } } */ /* { dg-final { scan-assembler-not {\tld4w\t} } } */ -/* { dg-final { scan-assembler-not {\tld1d\t} } } */ /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.b} 8 } } */ /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.h} 8 } } */ /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s} 8 } } */ -/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d} 8 } } */ /* { dg-final { scan-assembler-times {\tfaddv\th[0-9]+, p[0-7], z[0-9]+\.h} 4 } } */ /* { dg-final { scan-assembler-times {\tfaddv\ts[0-9]+, p[0-7], z[0-9]+\.s} 4 } } */ -/* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d} 4 } } */ /* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.b} 4 } } */ /* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.h} 6 } } */ /* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.s} 6 } } */ -/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.d} 6 } } */ /* { dg-final { scan-assembler-not {\tuqdec} } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly.c b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly.c new file mode 100644 index 00000000000..69c33192a7a --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly.c @@ -0,0 +1,43 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable -ffast-math -fno-vect-cost-model" } */ + +#include + +#define VEC_PERM(TYPE) \ +void __attribute__ ((noinline, noclone)) \ +vec_slp_##TYPE (TYPE *restrict a, TYPE *restrict b, int n) \ +{ \ + TYPE x0 = b[0]; \ + TYPE x1 = b[1]; \ + TYPE x2 = b[2]; \ + TYPE x3 = b[3]; \ + for (int i = 0; i < n; ++i) \ + { \ + x0 += a[i * 4]; \ + x1 += a[i * 4 + 1]; \ + x2 += a[i * 4 + 2]; \ + x3 += a[i * 4 + 3]; \ + } \ + b[0] = x0; \ + b[1] = x1; \ + b[2] = x2; \ + b[3] = x3; \ +} + +#define TEST_ALL(T) \ + T (int64_t) \ + T (uint64_t) \ + T (double) + +TEST_ALL (VEC_PERM) + +/* We can't use SLP for the 64-bit loops, since the number of reduction + results might be greater than the number of elements in the vector. */ +/* { dg-final { scan-assembler-times {\tld4d\t} 3 } } */ +/* { dg-final { scan-assembler-not {\tld1d\t} } } */ +/* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d} 8 } } */ +/* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d} 4 } } */ + +/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.d} 6 } } */ + +/* { dg-final { scan-assembler-not {\tuqdec} } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly_run.c b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly_run.c new file mode 100644 index 00000000000..fbf94329759 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_costly_run.c @@ -0,0 +1,5 @@ +/* { dg-do run { target aarch64_sve_hw } } */ +/* { dg-options "-O2 -ftree-vectorize -ffast-math -fno-vect-cost-model" } */ + +#define FILENAME "slp_7_costly.c" +#include "slp_7_run.c" diff --git a/gcc/testsuite/gcc.target/aarch64/sve/slp_7_run.c b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_run.c index 3cc090d5a14..7c0aa62d8d8 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/slp_7_run.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/slp_7_run.c @@ -1,7 +1,11 @@ /* { dg-do run { target aarch64_sve_hw } } */ /* { dg-options "-O2 -ftree-vectorize -ffast-math" } */ -#include "slp_7.c" +#ifndef FILENAME +#define FILENAME "slp_7.c" +#endif + +#include FILENAME #define N (54 * 4) diff --git a/gcc/testsuite/gcc.target/aarch64/sve/while_7.c b/gcc/testsuite/gcc.target/aarch64/sve/while_7.c index d5ffb66a142..a66a20d21f6 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/while_7.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/while_7.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable" } */ +/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable -fno-vect-cost-model" } */ #include diff --git a/gcc/testsuite/gcc.target/aarch64/sve/while_9.c b/gcc/testsuite/gcc.target/aarch64/sve/while_9.c index 9a8e5fe12fb..dd3f404ab39 100644 --- a/gcc/testsuite/gcc.target/aarch64/sve/while_9.c +++ b/gcc/testsuite/gcc.target/aarch64/sve/while_9.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable" } */ +/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable -fno-vect-cost-model" } */ #include -- 2.30.2