From patchwork Tue Feb 27 14:38:47 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: =?utf-8?q?Alex_Benn=C3=A9e?= X-Patchwork-Id: 129815 Delivered-To: patch@linaro.org Received: by 10.80.172.228 with SMTP id x91csp370972edc; Tue, 27 Feb 2018 06:55:51 -0800 (PST) X-Google-Smtp-Source: AG47ELsrh2opxnfE5E1q34hxGRWsf+WCRLC9ZKkxr8E27XU1vxTS1aCF61dY3b3cIivTEdGLCXUC X-Received: by 2002:a25:2d4a:: with SMTP id s10-v6mr9647735ybe.354.1519743351334; Tue, 27 Feb 2018 06:55:51 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1519743351; cv=none; d=google.com; s=arc-20160816; b=Gq3RjVEOJjMSW+Evbpk4tjlzer3PbdT490D4IP1DYuaWdcfzg0TtQE0InMa93j1vhv nq/S2OU/uJRrnxasBbHmwAsArLLOz+IkQ6c/O3is1Y6FvXlVCDpfsbUfBvyK1SCs/OeX gu+PuXRbI2VeMy01tAI/5bTQz40oXQs0RLX8uJxym8JIcgoeJTLBdD1Y2562Gni2RLV3 ST/PH3J3SqiZs1LOEsIc9gSngHZcH8VWVZTr2UXwWwMYxcviEO6n2ncb8YeCNn/iWpE3 sK6URYMeJmrxvaZOcN6t/WZqfcFUAD7tUNDs2OJn/Gjnl7DxUcRliMi3U1vCDU+NEjWb 4Iaw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:cc:list-subscribe:list-help:list-post:list-archive :list-unsubscribe:list-id:precedence:subject :content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:arc-authentication-results; bh=BgahG73tkqG0tC2b6+gKAAuvzzecpl1XrnbvPjhZc3Q=; b=ZIDAp01PCBYKfKh8MtT71W84KUR63APWlhChFGkpYuQ0F681S0abQykCyVkd7Uy5g5 cO1Jq5OuNY+sSnGayM0MSc8cfEiSCXa6bCPn58Wjwwerehc1mJMCy1d/c5VQdkYCIDXd MqmlJUvejOUFbHrhht1VP6qZHSxDGRX0qai6WxeauC/Zeh0KAvd0SfzKvSlchkfpWPmD bMrQe52nsDrF6A918iOxSmmBnGJK0a4CB2ME0DaNFsKAJOM9l6ZHWMstSfJACqQgRVEL rVrN7hPN+lEsSUUqNtqGhOc8BHuIymy8Qk2ort8gAxxC50dmwT1FDUXbrffhkxkIiHhD xlEg== ARC-Authentication-Results: i=1; mx.google.com; dkim=fail header.i=@linaro.org header.s=google header.b=HAsSJGDd; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 2001:4830:134:3::11 as permitted sender) smtp.mailfrom=qemu-devel-bounces+patch=linaro.org@nongnu.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=linaro.org Return-Path: Received: from lists.gnu.org (lists.gnu.org. [2001:4830:134:3::11]) by mx.google.com with ESMTPS id c188-v6si1969699ybb.461.2018.02.27.06.55.51 for (version=TLS1 cipher=AES128-SHA bits=128/128); Tue, 27 Feb 2018 06:55:51 -0800 (PST) Received-SPF: pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 2001:4830:134:3::11 as permitted sender) client-ip=2001:4830:134:3::11; Authentication-Results: mx.google.com; dkim=fail header.i=@linaro.org header.s=google header.b=HAsSJGDd; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 2001:4830:134:3::11 as permitted sender) smtp.mailfrom=qemu-devel-bounces+patch=linaro.org@nongnu.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=linaro.org Received: from localhost ([::1]:37774 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eqgfW-0000jx-JK for patch@linaro.org; Tue, 27 Feb 2018 09:55:50 -0500 Received: from eggs.gnu.org ([2001:4830:134:3::10]:58512) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eqgYD-0003gD-Jr for qemu-devel@nongnu.org; Tue, 27 Feb 2018 09:48:24 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eqgY5-0001XZ-Ns for qemu-devel@nongnu.org; Tue, 27 Feb 2018 09:48:17 -0500 Received: from mail-wr0-x242.google.com ([2a00:1450:400c:c0c::242]:40967) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1eqgY5-0001So-3f for qemu-devel@nongnu.org; Tue, 27 Feb 2018 09:48:09 -0500 Received: by mail-wr0-x242.google.com with SMTP id f14so25125914wre.8 for ; Tue, 27 Feb 2018 06:48:08 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=BgahG73tkqG0tC2b6+gKAAuvzzecpl1XrnbvPjhZc3Q=; b=HAsSJGDdkc+MgFYRwcjqGM99EZHmTgLZdYK0CvxgitBZmb61/ERtloO4YRf1HC1BBy WO8hwU9b8K0MwZ1etZhkSZoUSFW3EYBI1fMKyUSfi1qwTj1OWyutNg9YTfCPx286jWJW zQT9uFLDpRirUQXG13kHFhO4rFfQR2qf9OzyE= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=BgahG73tkqG0tC2b6+gKAAuvzzecpl1XrnbvPjhZc3Q=; b=cbF4x9/YYrr6kPSLBjxMx9tBC7GXjUALTQSDJmdVJvG3lruhMp3oT5eEIqFbZc2XbP Umfc11piWBawoLIJOK+M+AjlKfVhxaoullesHzACB0HOHJqm9wRiNbEaVf4Vsbk1tb7N jGegWNuyVA4XoIuVQ0ABBwYOgUratOLl4Fq0PBc8mRg+SaRWChlQcENdnAJ5zfD3oNlA +mpfD2l8JVHezQEFsJCqzOHxiGstNySAopCvbLmVb8C00ad1xsXrvsYyIMDCMsbHkqTR ig/7OGfaRB+i1Le1CkZzm/VCAnb5stRxloaG5djjY1e5EfhIOPWL3GwoeFM3MbSiHieB cOnw== X-Gm-Message-State: APf1xPASmJYwKH0/7ZopM3Hgkt26LuZGjw+DNg6KC/MzRh/hBdGSeUsZ +1GvTdBtnmWFvuq0Mefv9VPqDg== X-Received: by 10.223.139.68 with SMTP id v4mr13466126wra.23.1519742887815; Tue, 27 Feb 2018 06:48:07 -0800 (PST) Received: from zen.linaro.local ([81.128.185.34]) by smtp.gmail.com with ESMTPSA id u48sm11091448wrb.85.2018.02.27.06.48.01 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Tue, 27 Feb 2018 06:48:05 -0800 (PST) Received: from zen.linaroharston (localhost [127.0.0.1]) by zen.linaro.local (Postfix) with ESMTP id DC2C43E1647; Tue, 27 Feb 2018 14:38:54 +0000 (GMT) From: =?utf-8?q?Alex_Benn=C3=A9e?= To: qemu-arm@nongnu.org Date: Tue, 27 Feb 2018 14:38:47 +0000 Message-Id: <20180227143852.11175-27-alex.bennee@linaro.org> X-Mailer: git-send-email 2.15.1 In-Reply-To: <20180227143852.11175-1-alex.bennee@linaro.org> References: <20180227143852.11175-1-alex.bennee@linaro.org> MIME-Version: 1.0 X-detected-operating-system: by eggs.gnu.org: Genre and OS details not recognized. X-Received-From: 2a00:1450:400c:c0c::242 Subject: [Qemu-devel] [PATCH v4 26/31] arm/helper.c: re-factor rsqrte and add rsqrte_f16 X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: =?utf-8?q?Alex_Benn=C3=A9e?= , richard.henderson@linaro.org, qemu-devel@nongnu.org, Peter Maydell Errors-To: qemu-devel-bounces+patch=linaro.org@nongnu.org Sender: "Qemu-devel" Much like recpe the ARM ARM has simplified the pseudo code for the calculation which is done on a fixed point 9 bit integer maths. So while adding f16 we can also clean this up to be a little less heavy on the floating point and just return the fractional part and leave the calle's to do the final packing of the result. Signed-off-by: Alex Bennée Reviewed-by: Richard Henderson --- v2 - checkpatch fixes --- target/arm/helper.c | 221 ++++++++++++++++++++++++---------------------------- target/arm/helper.h | 1 + 2 files changed, 104 insertions(+), 118 deletions(-) -- 2.15.1 diff --git a/target/arm/helper.c b/target/arm/helper.c index e2d0ff0b4c..c82f63d440 100644 --- a/target/arm/helper.c +++ b/target/arm/helper.c @@ -11756,56 +11756,97 @@ float64 HELPER(recpe_f64)(float64 input, void *fpstp) /* The algorithm that must be used to calculate the estimate * is specified by the ARM ARM. */ -static float64 recip_sqrt_estimate(float64 a, float_status *real_fp_status) -{ - /* These calculations mustn't set any fp exception flags, - * so we use a local copy of the fp_status. - */ - float_status dummy_status = *real_fp_status; - float_status *s = &dummy_status; - float64 q; - int64_t q_int; - - if (float64_lt(a, float64_half, s)) { - /* range 0.25 <= a < 0.5 */ - - /* a in units of 1/512 rounded down */ - /* q0 = (int)(a * 512.0); */ - q = float64_mul(float64_512, a, s); - q_int = float64_to_int64_round_to_zero(q, s); - - /* reciprocal root r */ - /* r = 1.0 / sqrt(((double)q0 + 0.5) / 512.0); */ - q = int64_to_float64(q_int, s); - q = float64_add(q, float64_half, s); - q = float64_div(q, float64_512, s); - q = float64_sqrt(q, s); - q = float64_div(float64_one, q, s); + +static int do_recip_sqrt_estimate(int a) +{ + int b, estimate; + + assert(128 <= a && a < 512); + if (a < 256) { + a = a * 2 + 1; } else { - /* range 0.5 <= a < 1.0 */ + a = (a >> 1) << 1; + a = (a + 1) * 2; + } + b = 512; + while (a * (b + 1) * (b + 1) < (1 << 28)) { + b += 1; + } + estimate = (b + 1) / 2; + assert(256 <= estimate && estimate < 512); + + return estimate; +} + - /* a in units of 1/256 rounded down */ - /* q1 = (int)(a * 256.0); */ - q = float64_mul(float64_256, a, s); - int64_t q_int = float64_to_int64_round_to_zero(q, s); +static uint64_t recip_sqrt_estimate(int *exp , int exp_off, uint64_t frac) +{ + int estimate; + uint32_t scaled; - /* reciprocal root r */ - /* r = 1.0 /sqrt(((double)q1 + 0.5) / 256); */ - q = int64_to_float64(q_int, s); - q = float64_add(q, float64_half, s); - q = float64_div(q, float64_256, s); - q = float64_sqrt(q, s); - q = float64_div(float64_one, q, s); + if (*exp == 0) { + while (extract64(frac, 51, 1) == 0) { + frac = frac << 1; + *exp -= 1; + } + frac = extract64(frac, 0, 51) << 1; } - /* r in units of 1/256 rounded to nearest */ - /* s = (int)(256.0 * r + 0.5); */ - q = float64_mul(q, float64_256,s ); - q = float64_add(q, float64_half, s); - q_int = float64_to_int64_round_to_zero(q, s); + if (*exp & 1) { + /* scaled = UInt('01':fraction<51:45>) */ + scaled = deposit32(1 << 7, 0, 7, extract64(frac, 45, 7)); + } else { + /* scaled = UInt('1':fraction<51:44>) */ + scaled = deposit32(1 << 8, 0, 8, extract64(frac, 44, 8)); + } + estimate = do_recip_sqrt_estimate(scaled); - /* return (double)s / 256.0;*/ - return float64_div(int64_to_float64(q_int, s), float64_256, s); + *exp = (exp_off - *exp) / 2; + return extract64(estimate, 0, 8) << 44; +} + +float16 HELPER(rsqrte_f16)(float16 input, void *fpstp) +{ + float_status *s = fpstp; + float16 f16 = float16_squash_input_denormal(input, s); + uint16_t val = float16_val(f16); + bool f16_sign = float16_is_neg(f16); + int f16_exp = extract32(val, 10, 5); + uint16_t f16_frac = extract32(val, 0, 10); + uint64_t f64_frac; + + if (float16_is_any_nan(f16)) { + float16 nan = f16; + if (float16_is_signaling_nan(f16, s)) { + float_raise(float_flag_invalid, s); + nan = float16_maybe_silence_nan(f16, s); + } + if (s->default_nan_mode) { + nan = float16_default_nan(s); + } + return nan; + } else if (float16_is_zero(f16)) { + float_raise(float_flag_divbyzero, s); + return float16_set_sign(float16_infinity, f16_sign); + } else if (f16_sign) { + float_raise(float_flag_invalid, s); + return float16_default_nan(s); + } else if (float16_is_infinity(f16)) { + return float16_zero; + } + + /* Scale and normalize to a double-precision value between 0.25 and 1.0, + * preserving the parity of the exponent. */ + + f64_frac = ((uint64_t) f16_frac) << (52 - 10); + + f64_frac = recip_sqrt_estimate(&f16_exp, 44, f64_frac); + + /* result = sign : result_exp<4:0> : estimate<7:0> : Zeros(2) */ + val = deposit32(0, 15, 1, f16_sign); + val = deposit32(val, 10, 5, f16_exp); + val = deposit32(val, 2, 8, extract64(f64_frac, 52 - 8, 8)); + return make_float16(val); } float32 HELPER(rsqrte_f32)(float32 input, void *fpstp) @@ -11813,13 +11854,10 @@ float32 HELPER(rsqrte_f32)(float32 input, void *fpstp) float_status *s = fpstp; float32 f32 = float32_squash_input_denormal(input, s); uint32_t val = float32_val(f32); - uint32_t f32_sbit = 0x80000000 & val; - int32_t f32_exp = extract32(val, 23, 8); + uint32_t f32_sign = float32_is_neg(f32); + int f32_exp = extract32(val, 23, 8); uint32_t f32_frac = extract32(val, 0, 23); uint64_t f64_frac; - uint64_t val64; - int result_exp; - float64 f64; if (float32_is_any_nan(f32)) { float32 nan = f32; @@ -11845,32 +11883,13 @@ float32 HELPER(rsqrte_f32)(float32 input, void *fpstp) * preserving the parity of the exponent. */ f64_frac = ((uint64_t) f32_frac) << 29; - if (f32_exp == 0) { - while (extract64(f64_frac, 51, 1) == 0) { - f64_frac = f64_frac << 1; - f32_exp = f32_exp-1; - } - f64_frac = extract64(f64_frac, 0, 51) << 1; - } - - if (extract64(f32_exp, 0, 1) == 0) { - f64 = make_float64(((uint64_t) f32_sbit) << 32 - | (0x3feULL << 52) - | f64_frac); - } else { - f64 = make_float64(((uint64_t) f32_sbit) << 32 - | (0x3fdULL << 52) - | f64_frac); - } - result_exp = (380 - f32_exp) / 2; + f64_frac = recip_sqrt_estimate(&f32_exp, 380, f64_frac); - f64 = recip_sqrt_estimate(f64, s); - - val64 = float64_val(f64); - - val = ((result_exp & 0xff) << 23) - | ((val64 >> 29) & 0x7fffff); + /* result = sign : result_exp<4:0> : estimate<7:0> : Zeros(15) */ + val = deposit32(0, 31, 1, f32_sign); + val = deposit32(val, 23, 8, f32_exp); + val = deposit32(val, 15, 8, extract64(f64_frac, 52 - 8, 8)); return make_float32(val); } @@ -11879,11 +11898,9 @@ float64 HELPER(rsqrte_f64)(float64 input, void *fpstp) float_status *s = fpstp; float64 f64 = float64_squash_input_denormal(input, s); uint64_t val = float64_val(f64); - uint64_t f64_sbit = 0x8000000000000000ULL & val; - int64_t f64_exp = extract64(val, 52, 11); + bool f64_sign = float64_is_neg(f64); + int f64_exp = extract64(val, 52, 11); uint64_t f64_frac = extract64(val, 0, 52); - int64_t result_exp; - uint64_t result_frac; if (float64_is_any_nan(f64)) { float64 nan = f64; @@ -11905,36 +11922,13 @@ float64 HELPER(rsqrte_f64)(float64 input, void *fpstp) return float64_zero; } - /* Scale and normalize to a double-precision value between 0.25 and 1.0, - * preserving the parity of the exponent. */ - - if (f64_exp == 0) { - while (extract64(f64_frac, 51, 1) == 0) { - f64_frac = f64_frac << 1; - f64_exp = f64_exp - 1; - } - f64_frac = extract64(f64_frac, 0, 51) << 1; - } + f64_frac = recip_sqrt_estimate(&f64_exp, 3068, f64_frac); - if (extract64(f64_exp, 0, 1) == 0) { - f64 = make_float64(f64_sbit - | (0x3feULL << 52) - | f64_frac); - } else { - f64 = make_float64(f64_sbit - | (0x3fdULL << 52) - | f64_frac); - } - - result_exp = (3068 - f64_exp) / 2; - - f64 = recip_sqrt_estimate(f64, s); - - result_frac = extract64(float64_val(f64), 0, 52); - - return make_float64(f64_sbit | - ((result_exp & 0x7ff) << 52) | - result_frac); + /* result = sign : result_exp<4:0> : estimate<7:0> : Zeros(44) */ + val = deposit64(0, 61, 1, f64_sign); + val = deposit64(val, 52, 11, f64_exp); + val = deposit64(val, 44, 8, extract64(f64_frac, 52 - 8, 8)); + return make_float64(val); } uint32_t HELPER(recpe_u32)(uint32_t a, void *fpstp) @@ -11954,24 +11948,15 @@ uint32_t HELPER(recpe_u32)(uint32_t a, void *fpstp) uint32_t HELPER(rsqrte_u32)(uint32_t a, void *fpstp) { - float_status *fpst = fpstp; - float64 f64; + int estimate; if ((a & 0xc0000000) == 0) { return 0xffffffff; } - if (a & 0x80000000) { - f64 = make_float64((0x3feULL << 52) - | ((uint64_t)(a & 0x7fffffff) << 21)); - } else { /* bits 31-30 == '01' */ - f64 = make_float64((0x3fdULL << 52) - | ((uint64_t)(a & 0x3fffffff) << 22)); - } - - f64 = recip_sqrt_estimate(f64, fpst); + estimate = do_recip_sqrt_estimate(extract32(a, 23, 9)); - return 0x80000000 | ((float64_val(f64) >> 21) & 0x7fffffff); + return deposit32(0, 23, 9, estimate); } /* VFPv4 fused multiply-accumulate */ diff --git a/target/arm/helper.h b/target/arm/helper.h index 81d7baed6d..6dd8504ec3 100644 --- a/target/arm/helper.h +++ b/target/arm/helper.h @@ -195,6 +195,7 @@ DEF_HELPER_3(rsqrts_f32, f32, f32, f32, env) DEF_HELPER_FLAGS_2(recpe_f16, TCG_CALL_NO_RWG, f16, f16, ptr) DEF_HELPER_FLAGS_2(recpe_f32, TCG_CALL_NO_RWG, f32, f32, ptr) DEF_HELPER_FLAGS_2(recpe_f64, TCG_CALL_NO_RWG, f64, f64, ptr) +DEF_HELPER_FLAGS_2(rsqrte_f16, TCG_CALL_NO_RWG, f16, f16, ptr) DEF_HELPER_FLAGS_2(rsqrte_f32, TCG_CALL_NO_RWG, f32, f32, ptr) DEF_HELPER_FLAGS_2(rsqrte_f64, TCG_CALL_NO_RWG, f64, f64, ptr) DEF_HELPER_2(recpe_u32, i32, i32, ptr)