From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:35796) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1f0hFI-0007PD-8X for qemu-devel@nongnu.org; Tue, 27 Mar 2018 01:34:10 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1f0hFE-0005ON-NA for qemu-devel@nongnu.org; Tue, 27 Mar 2018 01:34:08 -0400 Received: from out5-smtp.messagingengine.com ([66.111.4.29]:48435) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1f0hFE-0005Nd-J6 for qemu-devel@nongnu.org; Tue, 27 Mar 2018 01:34:04 -0400 From: "Emilio G. Cota" Date: Tue, 27 Mar 2018 01:33:59 -0400 Message-Id: <1522128840-498-14-git-send-email-cota@braap.org> In-Reply-To: <1522128840-498-1-git-send-email-cota@braap.org> References: <1522128840-498-1-git-send-email-cota@braap.org> Subject: [Qemu-devel] [PATCH v2 13/14] hardfloat: support float32/64 comparison List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: Aurelien Jarno , Peter Maydell , =?UTF-8?q?Alex=20Benn=C3=A9e?= , Laurent Vivier , Richard Henderson , Paolo Bonzini , Mark Cave-Ayland Performance results for fp-bench run under aarch64-linux-user on an Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz host: - before: cmp-single: 34.23 MFlops cmp-double: 32.53 MFlops - after: cmp-single: 43.51 MFlops cmp-double: 41.23 MFlops Using float32/64_is_any_nan vs. isnan yields only up to a 2% perf difference, so I'm keeping for now a single implementation. This low sensitivity is most likely due to the soft-fp int64_to_float32/64 functions -- they take ~50% of execution time. They should be converted to hardfloat once there are test cases in fp-test for them. Signed-off-by: Emilio G. Cota --- fpu/softfloat.c | 69 +++++++++++++++++++++++++++++++++++++++++++++------------ 1 file changed, 55 insertions(+), 14 deletions(-) diff --git a/fpu/softfloat.c b/fpu/softfloat.c index ba7289b..2b86d73 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -2300,28 +2300,69 @@ static int compare_floats(FloatParts a, FloatParts b, bool is_quiet, } } -#define COMPARE(sz) \ -int float ## sz ## _compare(float ## sz a, float ## sz b, \ - float_status *s) \ -{ \ - FloatParts pa = float ## sz ## _unpack_canonical(a, s); \ - FloatParts pb = float ## sz ## _unpack_canonical(b, s); \ - return compare_floats(pa, pb, false, s); \ -} \ -int float ## sz ## _compare_quiet(float ## sz a, float ## sz b, \ - float_status *s) \ +#define COMPARE(attr, sz) \ +static int attr \ +soft_float ## sz ## _compare(float ## sz a, float ## sz b, \ + bool is_quiet, float_status *s) \ { \ FloatParts pa = float ## sz ## _unpack_canonical(a, s); \ FloatParts pb = float ## sz ## _unpack_canonical(b, s); \ - return compare_floats(pa, pb, true, s); \ + return compare_floats(pa, pb, is_quiet, s); \ } -COMPARE(16) -COMPARE(32) -COMPARE(64) +COMPARE(, 16) +COMPARE(__attribute__((noinline)), 32) +COMPARE(__attribute__((noinline)), 64) #undef COMPARE +int __attribute__((flatten)) +float16_compare(float16 a, float16 b, float_status *s) +{ + return soft_float16_compare(a, b, false, s); +} + +int __attribute__((flatten)) +float16_compare_quiet(float16 a, float16 b, float_status *s) +{ + return soft_float16_compare(a, b, true, s); +} + +#define GEN_FPU_COMPARE(name, soft_t, host_t) \ + static inline __attribute__((always_inline)) int \ + fpu_ ## name(soft_t a, soft_t b, bool is_quiet, 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 (unlikely(isnan(ha) || isnan(hb))) { \ + return soft_ ## name(a, b, is_quiet, s); \ + } \ + if (isgreater(ha, hb)) { \ + return float_relation_greater; \ + } \ + if (isless(ha, hb)) { \ + return float_relation_less; \ + } \ + return float_relation_equal; \ + } \ + \ + int name(soft_t a, soft_t b, float_status *s) \ + { \ + return fpu_ ## name(a, b, false, s); \ + } \ + \ + int name ## _quiet(soft_t a, soft_t b, float_status *s) \ + { \ + return fpu_ ## name(a, b, true, s); \ + } + +GEN_FPU_COMPARE(float32_compare, float32, float) +GEN_FPU_COMPARE(float64_compare, float64, double) +#undef GEN_FPU_COMPARE + /* Multiply A by 2 raised to the power N. */ static FloatParts scalbn_decomposed(FloatParts a, int n, float_status *s) { -- 2.7.4