diff options
Diffstat (limited to 'math/test/mathbench.c')
| -rw-r--r-- | math/test/mathbench.c | 229 |
1 files changed, 84 insertions, 145 deletions
diff --git a/math/test/mathbench.c b/math/test/mathbench.c index ed7e89bb7710..653c58fbc484 100644 --- a/math/test/mathbench.c +++ b/math/test/mathbench.c @@ -1,10 +1,23 @@ /* * Microbenchmark for math functions. * - * Copyright (c) 2018-2023, Arm Limited. + * Copyright (c) 2018-2024, Arm Limited. * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception */ +#if WANT_SVE_TESTS +# if __aarch64__ && __linux__ +# ifdef __clang__ +# pragma clang attribute push(__attribute__((target("sve"))), \ + apply_to = any(function)) +# else +# pragma GCC target("+sve") +# endif +# else +# error "SVE not supported - please disable WANT_SVE_TESTS" +# endif +#endif + #undef _GNU_SOURCE #define _GNU_SOURCE 1 #include <stdint.h> @@ -29,94 +42,6 @@ static float Af[N]; static long measurecount = MEASURE; static long itercount = ITER; -#ifdef __vpcs -#include <arm_neon.h> -typedef float64x2_t v_double; - -#define v_double_len() 2 - -static inline v_double -v_double_load (const double *p) -{ - return (v_double){p[0], p[1]}; -} - -static inline v_double -v_double_dup (double x) -{ - return (v_double){x, x}; -} - -typedef float32x4_t v_float; - -#define v_float_len() 4 - -static inline v_float -v_float_load (const float *p) -{ - return (v_float){p[0], p[1], p[2], p[3]}; -} - -static inline v_float -v_float_dup (float x) -{ - return (v_float){x, x, x, x}; -} -#else -/* dummy definitions to make things compile. */ -typedef double v_double; -typedef float v_float; -#define v_double_len(x) 1 -#define v_double_load(x) (x)[0] -#define v_double_dup(x) (x) -#define v_float_len(x) 1 -#define v_float_load(x) (x)[0] -#define v_float_dup(x) (x) - -#endif - -#if WANT_SVE_MATH -#include <arm_sve.h> -typedef svbool_t sv_bool; -typedef svfloat64_t sv_double; - -#define sv_double_len() svcntd() - -static inline sv_double -sv_double_load (const double *p) -{ - svbool_t pg = svptrue_b64(); - return svld1(pg, p); -} - -static inline sv_double -sv_double_dup (double x) -{ - return svdup_n_f64(x); -} - -typedef svfloat32_t sv_float; - -#define sv_float_len() svcntw() - -static inline sv_float -sv_float_load (const float *p) -{ - svbool_t pg = svptrue_b32(); - return svld1(pg, p); -} - -static inline sv_float -sv_float_dup (float x) -{ - return svdup_n_f32(x); -} -#else -/* dummy definitions to make things compile. */ -#define sv_double_len(x) 1 -#define sv_float_len(x) 1 -#endif - static double dummy (double x) { @@ -128,28 +53,28 @@ dummyf (float x) { return x; } -#ifdef __vpcs -__vpcs static v_double -__vn_dummy (v_double x) +#if __aarch64__ && __linux__ +__vpcs static float64x2_t +__vn_dummy (float64x2_t x) { return x; } -__vpcs static v_float -__vn_dummyf (v_float x) +__vpcs static float32x4_t +__vn_dummyf (float32x4_t x) { return x; } #endif -#if WANT_SVE_MATH -static sv_double -__sv_dummy (sv_double x, sv_bool pg) +#if WANT_SVE_TESTS +static svfloat64_t +__sv_dummy (svfloat64_t x, svbool_t pg) { return x; } -static sv_float -__sv_dummyf (sv_float x, sv_bool pg) +static svfloat32_t +__sv_dummyf (svfloat32_t x, svbool_t pg) { return x; } @@ -169,16 +94,17 @@ static const struct fun { double (*d) (double); float (*f) (float); -#ifdef __vpcs - __vpcs v_double (*vnd) (v_double); - __vpcs v_float (*vnf) (v_float); +#if __aarch64__ && __linux__ + __vpcs float64x2_t (*vnd) (float64x2_t); + __vpcs float32x4_t (*vnf) (float32x4_t); #endif -#if WANT_SVE_MATH - sv_double (*svd) (sv_double, sv_bool); - sv_float (*svf) (sv_float, sv_bool); +#if WANT_SVE_TESTS + svfloat64_t (*svd) (svfloat64_t, svbool_t); + svfloat32_t (*svf) (svfloat32_t, svbool_t); #endif } fun; } funtab[] = { +// clang-format off #define D(func, lo, hi) {#func, 'd', 0, lo, hi, {.d = func}}, #define F(func, lo, hi) {#func, 'f', 0, lo, hi, {.f = func}}, #define VND(func, lo, hi) {#func, 'd', 'n', lo, hi, {.vnd = func}}, @@ -187,11 +113,11 @@ static const struct fun #define SVF(func, lo, hi) {#func, 'f', 's', lo, hi, {.svf = func}}, D (dummy, 1.0, 2.0) F (dummyf, 1.0, 2.0) -#ifdef __vpcs +#if __aarch64__ && __linux__ VND (__vn_dummy, 1.0, 2.0) VNF (__vn_dummyf, 1.0, 2.0) #endif -#if WANT_SVE_MATH +#if WANT_SVE_TESTS SVD (__sv_dummy, 1.0, 2.0) SVF (__sv_dummyf, 1.0, 2.0) #endif @@ -203,6 +129,7 @@ SVF (__sv_dummyf, 1.0, 2.0) #undef VND #undef SVF #undef SVD + // clang-format on }; static void @@ -301,75 +228,77 @@ runf_latency (float f (float)) prev = f (Af[i] + prev * z); } -#ifdef __vpcs +#if __aarch64__ && __linux__ static void -run_vn_thruput (__vpcs v_double f (v_double)) +run_vn_thruput (__vpcs float64x2_t f (float64x2_t)) { - for (int i = 0; i < N; i += v_double_len ()) - f (v_double_load (A+i)); + for (int i = 0; i < N; i += 2) + f (vld1q_f64 (A + i)); } static void -runf_vn_thruput (__vpcs v_float f (v_float)) +runf_vn_thruput (__vpcs float32x4_t f (float32x4_t)) { - for (int i = 0; i < N; i += v_float_len ()) - f (v_float_load (Af+i)); + for (int i = 0; i < N; i += 4) + f (vld1q_f32 (Af + i)); } static void -run_vn_latency (__vpcs v_double f (v_double)) +run_vn_latency (__vpcs float64x2_t f (float64x2_t)) { volatile uint64x2_t vsel = (uint64x2_t) { 0, 0 }; uint64x2_t sel = vsel; - v_double prev = v_double_dup (0); - for (int i = 0; i < N; i += v_double_len ()) - prev = f (vbslq_f64 (sel, prev, v_double_load (A+i))); + float64x2_t prev = vdupq_n_f64 (0); + for (int i = 0; i < N; i += 2) + prev = f (vbslq_f64 (sel, prev, vld1q_f64 (A + i))); } static void -runf_vn_latency (__vpcs v_float f (v_float)) +runf_vn_latency (__vpcs float32x4_t f (float32x4_t)) { volatile uint32x4_t vsel = (uint32x4_t) { 0, 0, 0, 0 }; uint32x4_t sel = vsel; - v_float prev = v_float_dup (0); - for (int i = 0; i < N; i += v_float_len ()) - prev = f (vbslq_f32 (sel, prev, v_float_load (Af+i))); + float32x4_t prev = vdupq_n_f32 (0); + for (int i = 0; i < N; i += 4) + prev = f (vbslq_f32 (sel, prev, vld1q_f32 (Af + i))); } #endif -#if WANT_SVE_MATH +#if WANT_SVE_TESTS static void -run_sv_thruput (sv_double f (sv_double, sv_bool)) +run_sv_thruput (svfloat64_t f (svfloat64_t, svbool_t)) { - for (int i = 0; i < N; i += sv_double_len ()) - f (sv_double_load (A+i), svptrue_b64 ()); + for (int i = 0; i < N; i += svcntd ()) + f (svld1_f64 (svptrue_b64 (), A + i), svptrue_b64 ()); } static void -runf_sv_thruput (sv_float f (sv_float, sv_bool)) +runf_sv_thruput (svfloat32_t f (svfloat32_t, svbool_t)) { - for (int i = 0; i < N; i += sv_float_len ()) - f (sv_float_load (Af+i), svptrue_b32 ()); + for (int i = 0; i < N; i += svcntw ()) + f (svld1_f32 (svptrue_b32 (), Af + i), svptrue_b32 ()); } static void -run_sv_latency (sv_double f (sv_double, sv_bool)) +run_sv_latency (svfloat64_t f (svfloat64_t, svbool_t)) { - volatile sv_bool vsel = svptrue_b64 (); - sv_bool sel = vsel; - sv_double prev = sv_double_dup (0); - for (int i = 0; i < N; i += sv_double_len ()) - prev = f (svsel_f64 (sel, sv_double_load (A+i), prev), svptrue_b64 ()); + volatile svbool_t vsel = svptrue_b64 (); + svbool_t sel = vsel; + svfloat64_t prev = svdup_f64 (0); + for (int i = 0; i < N; i += svcntd ()) + prev = f (svsel_f64 (sel, svld1_f64 (svptrue_b64 (), A + i), prev), + svptrue_b64 ()); } static void -runf_sv_latency (sv_float f (sv_float, sv_bool)) +runf_sv_latency (svfloat32_t f (svfloat32_t, svbool_t)) { - volatile sv_bool vsel = svptrue_b32 (); - sv_bool sel = vsel; - sv_float prev = sv_float_dup (0); - for (int i = 0; i < N; i += sv_float_len ()) - prev = f (svsel_f32 (sel, sv_float_load (Af+i), prev), svptrue_b32 ()); + volatile svbool_t vsel = svptrue_b32 (); + svbool_t sel = vsel; + svfloat32_t prev = svdup_f32 (0); + for (int i = 0; i < N; i += svcntw ()) + prev = f (svsel_f32 (sel, svld1_f32 (svptrue_b32 (), Af + i), prev), + svptrue_b32 ()); } #endif @@ -377,7 +306,11 @@ static uint64_t tic (void) { struct timespec ts; +#if defined(_MSC_VER) + if (!timespec_get (&ts, TIME_UTC)) +#else if (clock_gettime (CLOCK_REALTIME, &ts)) +#endif abort (); return ts.tv_sec * 1000000000ULL + ts.tv_nsec; } @@ -405,9 +338,11 @@ bench1 (const struct fun *f, int type, double lo, double hi) int vlen = 1; if (f->vec == 'n') - vlen = f->prec == 'd' ? v_double_len() : v_float_len(); + vlen = f->prec == 'd' ? 2 : 4; +#if WANT_SVE_TESTS else if (f->vec == 's') - vlen = f->prec == 'd' ? sv_double_len() : sv_float_len(); + vlen = f->prec == 'd' ? svcntd () : svcntw (); +#endif if (f->prec == 'd' && type == 't' && f->vec == 0) TIMEIT (run_thruput, f->fun.d); @@ -417,7 +352,7 @@ bench1 (const struct fun *f, int type, double lo, double hi) TIMEIT (runf_thruput, f->fun.f); else if (f->prec == 'f' && type == 'l' && f->vec == 0) TIMEIT (runf_latency, f->fun.f); -#ifdef __vpcs +#if __aarch64__ && __linux__ else if (f->prec == 'd' && type == 't' && f->vec == 'n') TIMEIT (run_vn_thruput, f->fun.vnd); else if (f->prec == 'd' && type == 'l' && f->vec == 'n') @@ -427,7 +362,7 @@ bench1 (const struct fun *f, int type, double lo, double hi) else if (f->prec == 'f' && type == 'l' && f->vec == 'n') TIMEIT (runf_vn_latency, f->fun.vnf); #endif -#if WANT_SVE_MATH +#if WANT_SVE_TESTS else if (f->prec == 'd' && type == 't' && f->vec == 's') TIMEIT (run_sv_thruput, f->fun.svd); else if (f->prec == 'd' && type == 'l' && f->vec == 's') @@ -640,3 +575,7 @@ main (int argc, char *argv[]) } return 0; } + +#if __aarch64__ && __linux__ && WANT_SVE_TESTS && defined(__clang__) +# pragma clang attribute pop +#endif |