From patchwork Fri Jun 7 00:24:42 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Richard Henderson X-Patchwork-Id: 802306 Delivered-To: patch@linaro.org Received: by 2002:adf:f147:0:b0:35b:5a80:51b4 with SMTP id y7csp564549wro; Thu, 6 Jun 2024 17:25:29 -0700 (PDT) X-Forwarded-Encrypted: i=2; AJvYcCUx3U3nLCFxfvfCYSPgiX7YlLCie78VxXkxuvg8FrOZ4KwS6zQWAUsqDxbpEKOe/KpK7lo5uxa0Ml8HTjP4CeI5 X-Google-Smtp-Source: AGHT+IHAvZnIWhxMZ49kZSIEnj05TP7mwgHB6adO7XCEC2FuAjoZKJFdksgHZfbXdDtqile80wAm X-Received: by 2002:a05:622a:19a2:b0:43f:f291:11fa with SMTP id d75a77b69052e-44041ce3638mr13740341cf.41.1717719929526; Thu, 06 Jun 2024 17:25:29 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1717719929; cv=none; d=google.com; s=arc-20160816; b=ukMQDqQGH4irB1oScJOIJbCMXhuqMjx9oZR2MgG5OOTx7SrYdNmqFzFI99ErC1Q7sN DMigsIMsIp/sWmTJQfpXV3bgr6aEzVaE6cYt44GvCHwgI42t689bxctEM/RBeWl4gPPT QrVLvXv82GkaX8ZRdsFnl5ltJAKBpbHhdTmXklSjeBLPjY14P5X/nZaf1/itfaZoH1o3 YoCp55VnyR7As617gI13TRUItUfclYUYHIrJjpWK3cYsTeJKOXXugNxDWwjCILDdRRaB wAFi2/5OKrEz/91hOWWq7GMBsaGz0Rl/co1ISz3a2D0Ur87Pky3JYUg/jEO7xgV7CRga c8bA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:list-subscribe:list-help:list-post:list-archive :list-unsubscribe:list-id:precedence:content-transfer-encoding :mime-version:references:in-reply-to:message-id:date:subject:cc:to :from:dkim-signature; bh=we/cC6SV/OHCnqxYCDPX0iolNzRY4+fOCUveM9lfOU8=; fh=OexW7qhzt4rEPTYIA7PtuxuSScd1B5LZutrE59YueSY=; b=MSxyeokYqHGxvFqL97sq87BcFgQhwlB35reDccGyUli2lU5DT0utNjOXh0FFOmMEf5 K89+/Thfk7Rp0Jckufok1Ie5YqnedW114HbM2DVQoBFmpJm8KnMwWBas2pD/FLK9Nwe1 EAF+TmdONTiNvL5kHtANOWtaWILPjNTZb+ttLTUJQTGBhGRU2jI4gRnvrbk05fGvI/T3 adcCFuSvbMW7DO3BASGTe5tW9mvtILwuuqDFcFDbi38xecIas6tE17E298ZgqpjlfmId OplZPMdK+meo5+cICytw65eowOypkPuCqUusM6yMIgbTSw3wAdydwtlCDpiwIDVNfLQr /uAw==; dara=google.com ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@linaro.org header.s=google header.b=l3XdO79L; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) smtp.mailfrom="qemu-devel-bounces+patch=linaro.org@nongnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=linaro.org Return-Path: Received: from lists.gnu.org (lists.gnu.org. [209.51.188.17]) by mx.google.com with ESMTPS id d75a77b69052e-44038b81930si27906221cf.752.2024.06.06.17.25.29 for (version=TLS1_2 cipher=ECDHE-ECDSA-CHACHA20-POLY1305 bits=256/256); Thu, 06 Jun 2024 17:25:29 -0700 (PDT) Received-SPF: pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) client-ip=209.51.188.17; Authentication-Results: mx.google.com; dkim=pass header.i=@linaro.org header.s=google header.b=l3XdO79L; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) smtp.mailfrom="qemu-devel-bounces+patch=linaro.org@nongnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=linaro.org Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1sFNPY-0001EZ-RX; Thu, 06 Jun 2024 20:24:52 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1sFNPW-0001Dm-3o for qemu-devel@nongnu.org; Thu, 06 Jun 2024 20:24:50 -0400 Received: from mail-pf1-x42a.google.com ([2607:f8b0:4864:20::42a]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_128_GCM_SHA256:128) (Exim 4.90_1) (envelope-from ) id 1sFNPT-0006wK-OI for qemu-devel@nongnu.org; Thu, 06 Jun 2024 20:24:49 -0400 Received: by mail-pf1-x42a.google.com with SMTP id d2e1a72fcca58-704090c1204so569702b3a.1 for ; Thu, 06 Jun 2024 17:24:47 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; t=1717719886; x=1718324686; darn=nongnu.org; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:from:to:cc:subject:date :message-id:reply-to; bh=we/cC6SV/OHCnqxYCDPX0iolNzRY4+fOCUveM9lfOU8=; b=l3XdO79LCZv2gj/bpwphNkLzvr+wnlnzOy2QAGBB1eqzA+8Up5qfEbTrqK8IPVW7sM gcIJhOqeMnroUYJw8cjC2txsPgNz/UjajDlZFUp0N2RoPyljqRzDTtOa95MQTxzV0BIC c2Ixbe6kLm9K5LV22zOtruZuegU4EdcDqxkgVQh8x6iBjRPbcxNGvlQ7sWMsJKs0D6wE pPxFFYpifZewB6g/rlIacrvfTtm0JWXNFfRUEpLezUFh6qRyFHo9IaEOrLCJ5/s0ksVD 61biWOAAMBtmuxP7UIUlFB2zL0bTRY1VexOebQIvFykpyz4tk7FBAeD4QYwaycADPBeQ LVmw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1717719886; x=1718324686; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=we/cC6SV/OHCnqxYCDPX0iolNzRY4+fOCUveM9lfOU8=; b=pft5jwkI5cs++mhlT//QPdu0E6nxoGkP9rj5iFef0qfNvg+uDcwgG+ZJDjTYAaPqo3 QIxFaB0nLn86eDOQsjTMi3ieZI+87sTZn0S7r2157YAojamST/Dlz1bXwbiFVJv/X6V0 zsVivNifzGtnovq/y6tAVHVFdNAXVxE5sn2P4l6NU47BFcOdCLdDICY3CsDR4+vwTM+I EbtxlACVQ0jiRYuOCULYUSD/qkLGAbBJeE3iUEKx48u3R2nsPAERhpKjCmpz8xR9H9Pz nZWU8CC+O8Bsz+t97jpWgrpWy27pW9KhY4dkZyeXvUVknfHQ8mHiaj2fJjy3XfPi8HC9 PHtw== X-Gm-Message-State: AOJu0YwlGgxPicJ3CHBZp7sDZeAuPjFZstrc7pT2BsG2zvubM6RPCvrP /lCOH9Coh9NHr19WqCP7uDeNxF6F4+32X+Yx8CTGHqaml1d83jNoPoHeN3FunVCewwZ/IV/TR+I b X-Received: by 2002:a05:6a00:22d0:b0:6f3:e720:cead with SMTP id d2e1a72fcca58-7040c5e7099mr1303055b3a.5.1717719885682; Thu, 06 Jun 2024 17:24:45 -0700 (PDT) Received: from stoup.. ([75.147.178.105]) by smtp.gmail.com with ESMTPSA id 41be03b00d2f7-6e2fc776f24sm30985a12.25.2024.06.06.17.24.44 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 06 Jun 2024 17:24:45 -0700 (PDT) From: Richard Henderson To: qemu-devel@nongnu.org Cc: maobibo@loongson.cn, =?utf-8?q?Philippe_Mathieu-Daud=C3=A9?= Subject: [PATCH v2 1/2] util/bufferiszero: Split out host include files Date: Thu, 6 Jun 2024 17:24:42 -0700 Message-Id: <20240607002443.5820-2-richard.henderson@linaro.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240607002443.5820-1-richard.henderson@linaro.org> References: <20240607002443.5820-1-richard.henderson@linaro.org> MIME-Version: 1.0 Received-SPF: pass client-ip=2607:f8b0:4864:20::42a; envelope-from=richard.henderson@linaro.org; helo=mail-pf1-x42a.google.com X-Spam_score_int: -20 X-Spam_score: -2.1 X-Spam_bar: -- X-Spam_report: (-2.1 / 5.0 requ) BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_NONE=-0.0001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001, T_SCC_BODY_TEXT_LINE=-0.01 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-devel-bounces+patch=linaro.org@nongnu.org Sender: qemu-devel-bounces+patch=linaro.org@nongnu.org Split out host/bufferiszero.h.inc for x86, aarch64 and generic in order to avoid an overlong ifdef ladder. Reviewed-by: Philippe Mathieu-Daudé Signed-off-by: Richard Henderson --- util/bufferiszero.c | 191 +------------------ host/include/aarch64/host/bufferiszero.c.inc | 76 ++++++++ host/include/generic/host/bufferiszero.c.inc | 10 + host/include/i386/host/bufferiszero.c.inc | 124 ++++++++++++ host/include/x86_64/host/bufferiszero.c.inc | 1 + 5 files changed, 212 insertions(+), 190 deletions(-) create mode 100644 host/include/aarch64/host/bufferiszero.c.inc create mode 100644 host/include/generic/host/bufferiszero.c.inc create mode 100644 host/include/i386/host/bufferiszero.c.inc create mode 100644 host/include/x86_64/host/bufferiszero.c.inc diff --git a/util/bufferiszero.c b/util/bufferiszero.c index 11c080e02c..522146dab9 100644 --- a/util/bufferiszero.c +++ b/util/bufferiszero.c @@ -81,196 +81,7 @@ static bool buffer_is_zero_int_ge256(const void *buf, size_t len) return t == 0; } -#if defined(CONFIG_AVX2_OPT) || defined(__SSE2__) -#include - -/* Helper for preventing the compiler from reassociating - chains of binary vector operations. */ -#define SSE_REASSOC_BARRIER(vec0, vec1) asm("" : "+x"(vec0), "+x"(vec1)) - -/* Note that these vectorized functions may assume len >= 256. */ - -static bool __attribute__((target("sse2"))) -buffer_zero_sse2(const void *buf, size_t len) -{ - /* Unaligned loads at head/tail. */ - __m128i v = *(__m128i_u *)(buf); - __m128i w = *(__m128i_u *)(buf + len - 16); - /* Align head/tail to 16-byte boundaries. */ - const __m128i *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16); - const __m128i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16); - __m128i zero = { 0 }; - - /* Collect a partial block at tail end. */ - v |= e[-1]; w |= e[-2]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-3]; w |= e[-4]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-5]; w |= e[-6]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-7]; v |= w; - - /* - * Loop over complete 128-byte blocks. - * With the head and tail removed, e - p >= 14, so the loop - * must iterate at least once. - */ - do { - v = _mm_cmpeq_epi8(v, zero); - if (unlikely(_mm_movemask_epi8(v) != 0xFFFF)) { - return false; - } - v = p[0]; w = p[1]; - SSE_REASSOC_BARRIER(v, w); - v |= p[2]; w |= p[3]; - SSE_REASSOC_BARRIER(v, w); - v |= p[4]; w |= p[5]; - SSE_REASSOC_BARRIER(v, w); - v |= p[6]; w |= p[7]; - SSE_REASSOC_BARRIER(v, w); - v |= w; - p += 8; - } while (p < e - 7); - - return _mm_movemask_epi8(_mm_cmpeq_epi8(v, zero)) == 0xFFFF; -} - -#ifdef CONFIG_AVX2_OPT -static bool __attribute__((target("avx2"))) -buffer_zero_avx2(const void *buf, size_t len) -{ - /* Unaligned loads at head/tail. */ - __m256i v = *(__m256i_u *)(buf); - __m256i w = *(__m256i_u *)(buf + len - 32); - /* Align head/tail to 32-byte boundaries. */ - const __m256i *p = QEMU_ALIGN_PTR_DOWN(buf + 32, 32); - const __m256i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 32); - __m256i zero = { 0 }; - - /* Collect a partial block at tail end. */ - v |= e[-1]; w |= e[-2]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-3]; w |= e[-4]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-5]; w |= e[-6]; - SSE_REASSOC_BARRIER(v, w); - v |= e[-7]; v |= w; - - /* Loop over complete 256-byte blocks. */ - for (; p < e - 7; p += 8) { - /* PTEST is not profitable here. */ - v = _mm256_cmpeq_epi8(v, zero); - if (unlikely(_mm256_movemask_epi8(v) != 0xFFFFFFFF)) { - return false; - } - v = p[0]; w = p[1]; - SSE_REASSOC_BARRIER(v, w); - v |= p[2]; w |= p[3]; - SSE_REASSOC_BARRIER(v, w); - v |= p[4]; w |= p[5]; - SSE_REASSOC_BARRIER(v, w); - v |= p[6]; w |= p[7]; - SSE_REASSOC_BARRIER(v, w); - v |= w; - } - - return _mm256_movemask_epi8(_mm256_cmpeq_epi8(v, zero)) == 0xFFFFFFFF; -} -#endif /* CONFIG_AVX2_OPT */ - -static biz_accel_fn const accel_table[] = { - buffer_is_zero_int_ge256, - buffer_zero_sse2, -#ifdef CONFIG_AVX2_OPT - buffer_zero_avx2, -#endif -}; - -static unsigned best_accel(void) -{ -#ifdef CONFIG_AVX2_OPT - unsigned info = cpuinfo_init(); - - if (info & CPUINFO_AVX2) { - return 2; - } -#endif - return 1; -} - -#elif defined(__aarch64__) && defined(__ARM_NEON) -#include - -/* - * Helper for preventing the compiler from reassociating - * chains of binary vector operations. - */ -#define REASSOC_BARRIER(vec0, vec1) asm("" : "+w"(vec0), "+w"(vec1)) - -static bool buffer_is_zero_simd(const void *buf, size_t len) -{ - uint32x4_t t0, t1, t2, t3; - - /* Align head/tail to 16-byte boundaries. */ - const uint32x4_t *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16); - const uint32x4_t *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16); - - /* Unaligned loads at head/tail. */ - t0 = vld1q_u32(buf) | vld1q_u32(buf + len - 16); - - /* Collect a partial block at tail end. */ - t1 = e[-7] | e[-6]; - t2 = e[-5] | e[-4]; - t3 = e[-3] | e[-2]; - t0 |= e[-1]; - REASSOC_BARRIER(t0, t1); - REASSOC_BARRIER(t2, t3); - t0 |= t1; - t2 |= t3; - REASSOC_BARRIER(t0, t2); - t0 |= t2; - - /* - * Loop over complete 128-byte blocks. - * With the head and tail removed, e - p >= 14, so the loop - * must iterate at least once. - */ - do { - /* - * Reduce via UMAXV. Whatever the actual result, - * it will only be zero if all input bytes are zero. - */ - if (unlikely(vmaxvq_u32(t0) != 0)) { - return false; - } - - t0 = p[0] | p[1]; - t1 = p[2] | p[3]; - t2 = p[4] | p[5]; - t3 = p[6] | p[7]; - REASSOC_BARRIER(t0, t1); - REASSOC_BARRIER(t2, t3); - t0 |= t1; - t2 |= t3; - REASSOC_BARRIER(t0, t2); - t0 |= t2; - p += 8; - } while (p < e - 7); - - return vmaxvq_u32(t0) == 0; -} - -#define best_accel() 1 -static biz_accel_fn const accel_table[] = { - buffer_is_zero_int_ge256, - buffer_is_zero_simd, -}; -#else -#define best_accel() 0 -static biz_accel_fn const accel_table[1] = { - buffer_is_zero_int_ge256 -}; -#endif +#include "host/bufferiszero.c.inc" static biz_accel_fn buffer_is_zero_accel; static unsigned accel_index; diff --git a/host/include/aarch64/host/bufferiszero.c.inc b/host/include/aarch64/host/bufferiszero.c.inc new file mode 100644 index 0000000000..947ee7ca1f --- /dev/null +++ b/host/include/aarch64/host/bufferiszero.c.inc @@ -0,0 +1,76 @@ +/* + * SPDX-License-Identifier: GPL-2.0-or-later + * buffer_is_zero acceleration, aarch64 version. + */ + +#ifdef __ARM_NEON +#include + +/* + * Helper for preventing the compiler from reassociating + * chains of binary vector operations. + */ +#define REASSOC_BARRIER(vec0, vec1) asm("" : "+w"(vec0), "+w"(vec1)) + +static bool buffer_is_zero_simd(const void *buf, size_t len) +{ + uint32x4_t t0, t1, t2, t3; + + /* Align head/tail to 16-byte boundaries. */ + const uint32x4_t *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16); + const uint32x4_t *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16); + + /* Unaligned loads at head/tail. */ + t0 = vld1q_u32(buf) | vld1q_u32(buf + len - 16); + + /* Collect a partial block at tail end. */ + t1 = e[-7] | e[-6]; + t2 = e[-5] | e[-4]; + t3 = e[-3] | e[-2]; + t0 |= e[-1]; + REASSOC_BARRIER(t0, t1); + REASSOC_BARRIER(t2, t3); + t0 |= t1; + t2 |= t3; + REASSOC_BARRIER(t0, t2); + t0 |= t2; + + /* + * Loop over complete 128-byte blocks. + * With the head and tail removed, e - p >= 14, so the loop + * must iterate at least once. + */ + do { + /* + * Reduce via UMAXV. Whatever the actual result, + * it will only be zero if all input bytes are zero. + */ + if (unlikely(vmaxvq_u32(t0) != 0)) { + return false; + } + + t0 = p[0] | p[1]; + t1 = p[2] | p[3]; + t2 = p[4] | p[5]; + t3 = p[6] | p[7]; + REASSOC_BARRIER(t0, t1); + REASSOC_BARRIER(t2, t3); + t0 |= t1; + t2 |= t3; + REASSOC_BARRIER(t0, t2); + t0 |= t2; + p += 8; + } while (p < e - 7); + + return vmaxvq_u32(t0) == 0; +} + +static biz_accel_fn const accel_table[] = { + buffer_is_zero_int_ge256, + buffer_is_zero_simd, +}; + +#define best_accel() 1 +#else +# include "host/include/generic/host/bufferiszero.c.inc" +#endif diff --git a/host/include/generic/host/bufferiszero.c.inc b/host/include/generic/host/bufferiszero.c.inc new file mode 100644 index 0000000000..ea0875c24a --- /dev/null +++ b/host/include/generic/host/bufferiszero.c.inc @@ -0,0 +1,10 @@ +/* + * SPDX-License-Identifier: GPL-2.0-or-later + * buffer_is_zero acceleration, generic version. + */ + +static biz_accel_fn const accel_table[1] = { + buffer_is_zero_int_ge256 +}; + +#define best_accel() 0 diff --git a/host/include/i386/host/bufferiszero.c.inc b/host/include/i386/host/bufferiszero.c.inc new file mode 100644 index 0000000000..3b9605d806 --- /dev/null +++ b/host/include/i386/host/bufferiszero.c.inc @@ -0,0 +1,124 @@ +/* + * SPDX-License-Identifier: GPL-2.0-or-later + * buffer_is_zero acceleration, x86 version. + */ + +#if defined(CONFIG_AVX2_OPT) || defined(__SSE2__) +#include + +/* Helper for preventing the compiler from reassociating + chains of binary vector operations. */ +#define SSE_REASSOC_BARRIER(vec0, vec1) asm("" : "+x"(vec0), "+x"(vec1)) + +/* Note that these vectorized functions may assume len >= 256. */ + +static bool __attribute__((target("sse2"))) +buffer_zero_sse2(const void *buf, size_t len) +{ + /* Unaligned loads at head/tail. */ + __m128i v = *(__m128i_u *)(buf); + __m128i w = *(__m128i_u *)(buf + len - 16); + /* Align head/tail to 16-byte boundaries. */ + const __m128i *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16); + const __m128i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16); + __m128i zero = { 0 }; + + /* Collect a partial block at tail end. */ + v |= e[-1]; w |= e[-2]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-3]; w |= e[-4]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-5]; w |= e[-6]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-7]; v |= w; + + /* + * Loop over complete 128-byte blocks. + * With the head and tail removed, e - p >= 14, so the loop + * must iterate at least once. + */ + do { + v = _mm_cmpeq_epi8(v, zero); + if (unlikely(_mm_movemask_epi8(v) != 0xFFFF)) { + return false; + } + v = p[0]; w = p[1]; + SSE_REASSOC_BARRIER(v, w); + v |= p[2]; w |= p[3]; + SSE_REASSOC_BARRIER(v, w); + v |= p[4]; w |= p[5]; + SSE_REASSOC_BARRIER(v, w); + v |= p[6]; w |= p[7]; + SSE_REASSOC_BARRIER(v, w); + v |= w; + p += 8; + } while (p < e - 7); + + return _mm_movemask_epi8(_mm_cmpeq_epi8(v, zero)) == 0xFFFF; +} + +#ifdef CONFIG_AVX2_OPT +static bool __attribute__((target("avx2"))) +buffer_zero_avx2(const void *buf, size_t len) +{ + /* Unaligned loads at head/tail. */ + __m256i v = *(__m256i_u *)(buf); + __m256i w = *(__m256i_u *)(buf + len - 32); + /* Align head/tail to 32-byte boundaries. */ + const __m256i *p = QEMU_ALIGN_PTR_DOWN(buf + 32, 32); + const __m256i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 32); + __m256i zero = { 0 }; + + /* Collect a partial block at tail end. */ + v |= e[-1]; w |= e[-2]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-3]; w |= e[-4]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-5]; w |= e[-6]; + SSE_REASSOC_BARRIER(v, w); + v |= e[-7]; v |= w; + + /* Loop over complete 256-byte blocks. */ + for (; p < e - 7; p += 8) { + /* PTEST is not profitable here. */ + v = _mm256_cmpeq_epi8(v, zero); + if (unlikely(_mm256_movemask_epi8(v) != 0xFFFFFFFF)) { + return false; + } + v = p[0]; w = p[1]; + SSE_REASSOC_BARRIER(v, w); + v |= p[2]; w |= p[3]; + SSE_REASSOC_BARRIER(v, w); + v |= p[4]; w |= p[5]; + SSE_REASSOC_BARRIER(v, w); + v |= p[6]; w |= p[7]; + SSE_REASSOC_BARRIER(v, w); + v |= w; + } + + return _mm256_movemask_epi8(_mm256_cmpeq_epi8(v, zero)) == 0xFFFFFFFF; +} +#endif /* CONFIG_AVX2_OPT */ + +static biz_accel_fn const accel_table[] = { + buffer_is_zero_int_ge256, + buffer_zero_sse2, +#ifdef CONFIG_AVX2_OPT + buffer_zero_avx2, +#endif +}; + +static unsigned best_accel(void) +{ +#ifdef CONFIG_AVX2_OPT + unsigned info = cpuinfo_init(); + if (info & CPUINFO_AVX2) { + return 2; + } +#endif + return 1; +} + +#else +# include "host/include/generic/host/bufferiszero.c.inc" +#endif diff --git a/host/include/x86_64/host/bufferiszero.c.inc b/host/include/x86_64/host/bufferiszero.c.inc new file mode 100644 index 0000000000..1d3f1fd6f5 --- /dev/null +++ b/host/include/x86_64/host/bufferiszero.c.inc @@ -0,0 +1 @@ +#include "host/include/i386/host/bufferiszero.c.inc" From patchwork Fri Jun 7 00:24:43 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Richard Henderson X-Patchwork-Id: 802304 Delivered-To: patch@linaro.org Received: by 2002:adf:f147:0:b0:35b:5a80:51b4 with SMTP id y7csp564547wro; Thu, 6 Jun 2024 17:25:29 -0700 (PDT) X-Forwarded-Encrypted: i=2; AJvYcCUMBOTeWTYRPzCa6cWQ5JHX1qS4bmEjPGVuPNkksKB1uGTifhLBW+U84/ZYLpZ8Y5DOvliYLD60IoIdx+fNCXOj X-Google-Smtp-Source: AGHT+IGNT7EtkKgRlp2MRlXA4yoEjItEpqYEZN52FNWVRAkkQ4rRcMe9+4U/vJ2nkwwuPFG7S1bK X-Received: by 2002:a05:620a:2484:b0:792:99d0:646a with SMTP id af79cd13be357-7953c4ab3c6mr122813385a.49.1717719929525; Thu, 06 Jun 2024 17:25:29 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1717719929; cv=none; d=google.com; s=arc-20160816; b=bZJ+rPIJc8N+udnrgDl+lq+3f/kZoNVtoXP++2MmRzY4OTWxTDNZ8K8MPQ0w4BX4gE DEDPAGD4YSm+JORIk4gQYPnWR9KLP23/Thf+e3nbFwR3ehi8x+qkn9mivFh+WsuhlMmp vFHtuyhQGrgFnhx/5UD87mIpm6SfCA7xSplOa+SGilPaa++Ffy6ylXpAjDn1LNtfmGdv jrYZ+BWhCN+2dq2tXw9S7s77Wbm7K1BdCTm+oKJSBgWuj+gXO1dSKMefjFGq9Dw1Uho5 T25QRjLdRV7806wWZ4MMfBVdyVRcwdQx8YqUTqT1hJBqPGx5p5iBwwXNKycuYnuoW3Yi mO3g== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:list-subscribe:list-help:list-post:list-archive :list-unsubscribe:list-id:precedence:content-transfer-encoding :mime-version:references:in-reply-to:message-id:date:subject:cc:to :from:dkim-signature; bh=4I3vf7TYIsB5RvqOtxeYyi1Lfw6+r6PiX+cBf5g4qW4=; fh=A3uSdaROzARb955zJUdRDVbI3bPP36A1SWP1OMRQfQw=; b=Mo7h/cr8bpevSYAITZvgMv3GW7fPFVzC/D+vd+TRwKncqIEPUGuQfv8g7gckt04nZK 6dxWGIAD3TsBDY8lMEt+kt7Km0vEFJ8aIy3ZQgz8i30TvhU+hJq850e4uv4qC8bjHn9I LJNbgRcvof9LiQPLxv/94SNFfFukYwhPZn/YjK60Ory1xcbOHvZ4TEa6Cm/toHiHbxVH 855zLs+HOFHNfLHFwgJhMqe/DWaAk6M/eUM5AdlppMAQiWjJfclAGAdnaeBlz6EVjJTx sA5PUNRMY5JW1zbl9/6Wp2rJLMab2aXLrcS86Z7ll/MoQHypl1/XbFwiSOFTnESTilND 5iIw==; dara=google.com ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@linaro.org header.s=google header.b="w4R/JBZ8"; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) smtp.mailfrom="qemu-devel-bounces+patch=linaro.org@nongnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=linaro.org Return-Path: Received: from lists.gnu.org (lists.gnu.org. [209.51.188.17]) by mx.google.com with ESMTPS id d75a77b69052e-44038b666b5si28033341cf.623.2024.06.06.17.25.29 for (version=TLS1_2 cipher=ECDHE-ECDSA-CHACHA20-POLY1305 bits=256/256); Thu, 06 Jun 2024 17:25:29 -0700 (PDT) Received-SPF: pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) client-ip=209.51.188.17; Authentication-Results: mx.google.com; dkim=pass header.i=@linaro.org header.s=google header.b="w4R/JBZ8"; spf=pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 209.51.188.17 as permitted sender) smtp.mailfrom="qemu-devel-bounces+patch=linaro.org@nongnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=linaro.org Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1sFNPW-0001E5-Vu; Thu, 06 Jun 2024 20:24:51 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1sFNPW-0001Dl-2o for qemu-devel@nongnu.org; Thu, 06 Jun 2024 20:24:50 -0400 Received: from mail-oo1-xc2d.google.com ([2607:f8b0:4864:20::c2d]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_128_GCM_SHA256:128) (Exim 4.90_1) (envelope-from ) id 1sFNPU-0006wS-BK for qemu-devel@nongnu.org; Thu, 06 Jun 2024 20:24:49 -0400 Received: by mail-oo1-xc2d.google.com with SMTP id 006d021491bc7-5b9778bb7c8so830818eaf.3 for ; Thu, 06 Jun 2024 17:24:47 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; t=1717719887; x=1718324687; darn=nongnu.org; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:from:to:cc:subject:date :message-id:reply-to; bh=4I3vf7TYIsB5RvqOtxeYyi1Lfw6+r6PiX+cBf5g4qW4=; b=w4R/JBZ88salW1IBmjWTzyOHm4U4MzJ5Pux+F5v58eqLc74dB81VTzrKM+F7wxW9p/ hZIYNK/H1m707SbW8IYGs0OUqf4Cuhwkw+Lv/XZOrfj7sWir6jmMBbfOGNSK/pYVw3I9 gDiXogjcnrX9I8eIiJ6MB8iCR0fLZcMZWFK0dnLR2WxiASoZs8n4osw36yP4ypuPPnFB 9A/JV5NOWbVcC2FMhQpyb/FF3vhGts6s2L4ZpdmSzUu8MLYTd77Qq7FGscRNNbBnhJAF fv2Gx3ZwKG5jpVcKyuOaFNM/+hQgMr9OPYaCQkXrbFeZ6E/47oIYo/9gGQktovgocKGZ Q/bQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1717719887; x=1718324687; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=4I3vf7TYIsB5RvqOtxeYyi1Lfw6+r6PiX+cBf5g4qW4=; b=qBwovJwL2u3zPJAeBIa+cO1BNGRFoMhwAfskwLA1fKfYOAxkwVEsOCHAC698hyidfm Mal8hrXm1usNW5cBQMKCtozZuY4/Yqty8xfGSzaFvxMjDrh3f0N6t38K+ExACiNx4xn5 gmDvgdK3uYYA0zGIdNLH3rSUBrxpOx3Up3w3I2J5i53Gyb+pYFbPcFQcfgHObbYvpb6S eOnwXA/Ble+9yMKpUfBSyQ+9n2k5o+BdzS6kwfw0cLD/6KCfHdCjUb+M4Odw0Q2dp3xP 7u21RSgoZi/doOQSHpAf1WwCDc4kL/YWgMFUpCF/S4a+nI/NQqyJiQci58jIqgs7lgZ0 RwgA== X-Gm-Message-State: AOJu0YwfxvmhQZEm3eBDNESJLzDEwUGy7iamkLwNW+a06uJNoRohKcIB cvlPFXrPsfgvhbr0Hd7V+ODMKBzdsOcYaj/bhxwRij/fVRfMGs5TtUJCZ9EX983R/tloJYONO0/ t X-Received: by 2002:a05:6359:5fa3:b0:19f:17ed:1f1a with SMTP id e5c5f4694b2df-19f1ff835d1mr141804455d.22.1717719886643; Thu, 06 Jun 2024 17:24:46 -0700 (PDT) Received: from stoup.. ([75.147.178.105]) by smtp.gmail.com with ESMTPSA id 41be03b00d2f7-6e2fc776f24sm30985a12.25.2024.06.06.17.24.45 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 06 Jun 2024 17:24:46 -0700 (PDT) From: Richard Henderson To: qemu-devel@nongnu.org Cc: maobibo@loongson.cn Subject: [PATCH v2 2/2] util/bufferiszero: Add loongarch64 vector acceleration Date: Thu, 6 Jun 2024 17:24:43 -0700 Message-Id: <20240607002443.5820-3-richard.henderson@linaro.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240607002443.5820-1-richard.henderson@linaro.org> References: <20240607002443.5820-1-richard.henderson@linaro.org> MIME-Version: 1.0 Received-SPF: pass client-ip=2607:f8b0:4864:20::c2d; envelope-from=richard.henderson@linaro.org; helo=mail-oo1-xc2d.google.com X-Spam_score_int: -20 X-Spam_score: -2.1 X-Spam_bar: -- X-Spam_report: (-2.1 / 5.0 requ) BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_NONE=-0.0001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001, T_SCC_BODY_TEXT_LINE=-0.01 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-devel-bounces+patch=linaro.org@nongnu.org Sender: qemu-devel-bounces+patch=linaro.org@nongnu.org Use inline assembly because no release compiler allows per-function selection of the ISA. Signed-off-by: Richard Henderson Tested-by: Bibo Mao --- .../loongarch64/host/bufferiszero.c.inc | 143 ++++++++++++++++++ 1 file changed, 143 insertions(+) create mode 100644 host/include/loongarch64/host/bufferiszero.c.inc diff --git a/host/include/loongarch64/host/bufferiszero.c.inc b/host/include/loongarch64/host/bufferiszero.c.inc new file mode 100644 index 0000000000..69891eac80 --- /dev/null +++ b/host/include/loongarch64/host/bufferiszero.c.inc @@ -0,0 +1,143 @@ +/* + * SPDX-License-Identifier: GPL-2.0-or-later + * buffer_is_zero acceleration, loongarch64 version. + */ + +/* + * Builtins for LSX and LASX are introduced by gcc 14 and llvm 18, + * but as yet neither has support for attribute target, so neither + * is able to enable the optimization without globally enabling + * vector support. Since we want runtime detection, use assembly. + */ + +static bool buffer_is_zero_lsx(const void *buf, size_t len) +{ + const void *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16); + const void *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16) - (7 * 16); + const void *l = buf + len; + bool ret; + + asm("vld $vr0,%2,0\n\t" /* first: buf + 0 */ + "vld $vr1,%4,-16\n\t" /* last: buf + len - 16 */ + "vld $vr2,%3,0\n\t" /* e[0] */ + "vld $vr3,%3,16\n\t" /* e[1] */ + "vld $vr4,%3,32\n\t" /* e[2] */ + "vld $vr5,%3,48\n\t" /* e[3] */ + "vld $vr6,%3,64\n\t" /* e[4] */ + "vld $vr7,%3,80\n\t" /* e[5] */ + "vld $vr8,%3,96\n\t" /* e[6] */ + "vor.v $vr0,$vr0,$vr1\n\t" + "vor.v $vr2,$vr2,$vr3\n\t" + "vor.v $vr4,$vr4,$vr5\n\t" + "vor.v $vr6,$vr6,$vr7\n\t" + "vor.v $vr0,$vr0,$vr2\n\t" + "vor.v $vr4,$vr4,$vr6\n\t" + "vor.v $vr0,$vr0,$vr4\n\t" + "vor.v $vr0,$vr0,$vr8\n\t" + "or %0,$r0,$r0\n" /* prepare return false */ + "1:\n\t" + "vsetnez.v $fcc0,$vr0\n\t" + "bcnez $fcc0,2f\n\t" + "vld $vr0,%1,0\n\t" /* p[0] */ + "vld $vr1,%1,16\n\t" /* p[1] */ + "vld $vr2,%1,32\n\t" /* p[2] */ + "vld $vr3,%1,48\n\t" /* p[3] */ + "vld $vr4,%1,64\n\t" /* p[4] */ + "vld $vr5,%1,80\n\t" /* p[5] */ + "vld $vr6,%1,96\n\t" /* p[6] */ + "vld $vr7,%1,112\n\t" /* p[7] */ + "addi.d %1,%1,128\n\t" + "vor.v $vr0,$vr0,$vr1\n\t" + "vor.v $vr2,$vr2,$vr3\n\t" + "vor.v $vr4,$vr4,$vr5\n\t" + "vor.v $vr6,$vr6,$vr7\n\t" + "vor.v $vr0,$vr0,$vr2\n\t" + "vor.v $vr4,$vr4,$vr6\n\t" + "vor.v $vr0,$vr0,$vr4\n\t" + "bltu %1,%3,1b\n\t" + "vsetnez.v $fcc0,$vr0\n\t" + "bcnez $fcc0,2f\n\t" + "ori %0,$r0,1\n" + "2:" + : "=&r"(ret), "+r"(p) + : "r"(buf), "r"(e), "r"(l) + : "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", "f8", "fcc0"); + + return ret; +} + +static bool buffer_is_zero_lasx(const void *buf, size_t len) +{ + const void *p = QEMU_ALIGN_PTR_DOWN(buf + 32, 32); + const void *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 32) - (7 * 32); + const void *l = buf + len; + bool ret; + + asm("xvld $xr0,%2,0\n\t" /* first: buf + 0 */ + "xvld $xr1,%4,-32\n\t" /* last: buf + len - 32 */ + "xvld $xr2,%3,0\n\t" /* e[0] */ + "xvld $xr3,%3,32\n\t" /* e[1] */ + "xvld $xr4,%3,64\n\t" /* e[2] */ + "xvld $xr5,%3,96\n\t" /* e[3] */ + "xvld $xr6,%3,128\n\t" /* e[4] */ + "xvld $xr7,%3,160\n\t" /* e[5] */ + "xvld $xr8,%3,192\n\t" /* e[6] */ + "xvor.v $xr0,$xr0,$xr1\n\t" + "xvor.v $xr2,$xr2,$xr3\n\t" + "xvor.v $xr4,$xr4,$xr5\n\t" + "xvor.v $xr6,$xr6,$xr7\n\t" + "xvor.v $xr0,$xr0,$xr2\n\t" + "xvor.v $xr4,$xr4,$xr6\n\t" + "xvor.v $xr0,$xr0,$xr4\n\t" + "xvor.v $xr0,$xr0,$xr8\n\t" + "or %0,$r0,$r0\n\t" /* prepare return false */ + "bgeu %1,%3,2f\n" + "1:\n\t" + "xvsetnez.v $fcc0,$xr0\n\t" + "bcnez $fcc0,3f\n\t" + "xvld $xr0,%1,0\n\t" /* p[0] */ + "xvld $xr1,%1,32\n\t" /* p[1] */ + "xvld $xr2,%1,64\n\t" /* p[2] */ + "xvld $xr3,%1,96\n\t" /* p[3] */ + "xvld $xr4,%1,128\n\t" /* p[4] */ + "xvld $xr5,%1,160\n\t" /* p[5] */ + "xvld $xr6,%1,192\n\t" /* p[6] */ + "xvld $xr7,%1,224\n\t" /* p[7] */ + "addi.d %1,%1,256\n\t" + "xvor.v $xr0,$xr0,$xr1\n\t" + "xvor.v $xr2,$xr2,$xr3\n\t" + "xvor.v $xr4,$xr4,$xr5\n\t" + "xvor.v $xr6,$xr6,$xr7\n\t" + "xvor.v $xr0,$xr0,$xr2\n\t" + "xvor.v $xr4,$xr4,$xr6\n\t" + "xvor.v $xr0,$xr0,$xr4\n\t" + "bltu %1,%3,1b\n" + "2:\n\t" + "xvsetnez.v $fcc0,$xr0\n\t" + "bcnez $fcc0,3f\n\t" + "ori %0,$r0,1\n" + "3:" + : "=&r"(ret), "+r"(p) + : "r"(buf), "r"(e), "r"(l) + : "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", "f8", "fcc0"); + + return ret; +} + +static biz_accel_fn const accel_table[] = { + buffer_is_zero_int_ge256, + buffer_is_zero_lsx, + buffer_is_zero_lasx, +}; + +static unsigned best_accel(void) +{ + unsigned info = cpuinfo_init(); + if (info & CPUINFO_LASX) { + return 2; + } + if (info & CPUINFO_LSX) { + return 1; + } + return 0; +}