Performance results for fp-bench run under aarch64-linux-user on an Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz host:
- before: mul-single: 88.37 MFlops mul-double: 85.55 MFlops - after: mul-single: 115.06 MFlops mul-double: 124.67 MFlops - w/ both using float32/64_is_normal etc.: mul-single: 113.49 MFlops mul-double: 113.46 MFlops - w/ both using fpclassify etc.: mul-single: 105.70 MFlops mul-double: 127.69 MFlops Signed-off-by: Emilio G. Cota <c...@braap.org> --- fpu/softfloat.c | 77 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 73 insertions(+), 4 deletions(-) diff --git a/fpu/softfloat.c b/fpu/softfloat.c index e0ab0ca..9739a86 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -1044,8 +1044,8 @@ float16 __attribute__((flatten)) float16_mul(float16 a, float16 b, return float16_round_pack_canonical(pr, status); } -float32 __attribute__((flatten)) float32_mul(float32 a, float32 b, - float_status *status) +static float32 __attribute__((flatten, noinline)) +soft_float32_mul(float32 a, float32 b, float_status *status) { FloatParts pa = float32_unpack_canonical(a, status); FloatParts pb = float32_unpack_canonical(b, status); @@ -1054,8 +1054,8 @@ float32 __attribute__((flatten)) float32_mul(float32 a, float32 b, return float32_round_pack_canonical(pr, status); } -float64 __attribute__((flatten)) float64_mul(float64 a, float64 b, - float_status *status) +static float64 __attribute__((flatten, noinline)) +soft_float64_mul(float64 a, float64 b, float_status *status) { FloatParts pa = float64_unpack_canonical(a, status); FloatParts pb = float64_unpack_canonical(b, status); @@ -1064,6 +1064,75 @@ float64 __attribute__((flatten)) float64_mul(float64 a, float64 b, return float64_round_pack_canonical(pr, status); } +#define GEN_FPU_MUL(name, soft_t, host_t, host_abs_func, min_normal) \ + soft_t name(soft_t a, soft_t b, float_status *s) \ + { \ + soft_t ## _input_flush2(&a, &b, s); \ + if (likely((soft_t ## _is_normal(a) || soft_t ## _is_zero(a)) && \ + (soft_t ## _is_normal(b) || soft_t ## _is_zero(b)) && \ + s->float_exception_flags & float_flag_inexact && \ + s->float_rounding_mode == float_round_nearest_even)) { \ + if (soft_t ## _is_zero(a) || soft_t ## _is_zero(b)) { \ + bool signbit = soft_t ## _is_neg(a) ^ soft_t ## _is_neg(b); \ + \ + return soft_t ## _set_sign(soft_t ## _zero, signbit); \ + } else { \ + host_t ha = soft_t ## _to_ ## host_t(a); \ + host_t hb = soft_t ## _to_ ## host_t(b); \ + host_t hr = ha * hb; \ + soft_t r = host_t ## _to_ ## soft_t(hr); \ + \ + if (unlikely(soft_t ## _is_infinity(r))) { \ + s->float_exception_flags |= float_flag_overflow; \ + } else if (unlikely(host_abs_func(hr) <= min_normal)) { \ + goto soft; \ + } \ + return r; \ + } \ + } \ + soft: \ + return soft_ ## soft_t ## _mul(a, b, s); \ + } + +GEN_FPU_MUL(float32_mul, float32, float, fabsf, FLT_MIN) +#undef GEN_FPU_MUL + +#define GEN_FPU_MUL(name, soft_t, host_t, host_abs_func, min_normal) \ + soft_t name(soft_t a, soft_t b, float_status *s) \ + { \ + host_t ha, hb; \ + \ + soft_t ## _input_flush2(&a, &b, s); \ + ha = soft_t ## _to_ ## host_t(a); \ + hb = soft_t ## _to_ ## host_t(b); \ + if (likely((fpclassify(ha) == FP_NORMAL || \ + fpclassify(ha) == FP_ZERO) && \ + (fpclassify(hb) == FP_NORMAL || \ + fpclassify(hb) == FP_ZERO) && \ + s->float_exception_flags & float_flag_inexact && \ + s->float_rounding_mode == float_round_nearest_even)) { \ + if (soft_t ## _is_zero(a) || soft_t ## _is_zero(b)) { \ + bool signbit = soft_t ## _is_neg(a) ^ soft_t ## _is_neg(b); \ + \ + return soft_t ## _set_sign(soft_t ## _zero, signbit); \ + } else { \ + host_t hr = ha * hb; \ + \ + if (unlikely(isinf(hr))) { \ + s->float_exception_flags |= float_flag_overflow; \ + } else if (unlikely(host_abs_func(hr) <= min_normal)) { \ + goto soft; \ + } \ + return host_t ## _to_ ## soft_t(hr); \ + } \ + } \ + soft: \ + return soft_ ## soft_t ## _mul(a, b, s); \ + } + +GEN_FPU_MUL(float64_mul, float64, double, fabs, DBL_MIN) +#undef GEN_FPU_MUL + /* * Returns the result of multiplying the floating-point values `a' and * `b' then adding 'c', with no intermediate rounding step after the -- 2.7.4