From patchwork Tue Apr 30 19:42:46 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Richard Henderson X-Patchwork-Id: 793366 Delivered-To: patch@linaro.org Received: by 2002:a05:6000:54f:b0:34d:5089:5a9e with SMTP id b15csp437994wrf; Tue, 30 Apr 2024 12:44:09 -0700 (PDT) X-Forwarded-Encrypted: i=2; AJvYcCU/f6pBuIBOckIAW41LhVIDzM66pso6bj14tJyqEbjZvVdTVGuSkHcweVF9t53I36TIq0BK4L21jRlKYRo6MGds X-Google-Smtp-Source: AGHT+IGU185Rh9LAUhhlyYPPJnmR+xqsJy64DY7OXiMKnTWOTws83qvzf0ZBmIeQ5asFUCdV2T0e X-Received: by 2002:a05:6122:3694:b0:4d4:2931:7d4d with SMTP id ec20-20020a056122369400b004d429317d4dmr810886vkb.5.1714506248623; Tue, 30 Apr 2024 12:44:08 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1714506248; cv=none; d=google.com; s=arc-20160816; b=YcniiABho4LPucWFzbbXGuorsyz/rcRpW3XjYqxpPCgJbmAQGb1NtPdrF2u9RCIZWK 9p91vWJFTaG76u1vBr+Yj0kr7L+kAp5c2jkZN00CJQHFxavFtmup3hScP3lyLZwaW1vT wjczHm+vXDsBQqSk/6fclxezpbYO9z/1Q8888lV2DzqYOKEvMzZy0rKUL+O3m1vfCZmw 24I4BFSdK1jR9aygnmP9JuHfGfwdgQw1UNkr3vmBQbm4zQLMLdwRn+VomkDoyOXOH2hz z/lLvGMT2Yk1uS3BVxvWz7SqhjPpkalUTGjcReW0ZiuALx7s8FMHuY1i/1ZVFa7sPnow X2PA== 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=Aco89HGhDUqxtzZSl7gQEXCIzX7CLhr3rgmAxnnD0GI=; fh=QtYERfLY+/6lXIVoMxqjAcWASzkMPfeQc3Ur4v8xQcU=; b=QwK9YkESL4HRX9rheFSppXKEJA0dVYYjFdIuBABmeegWkzuvtICrGrkr5+tKM4sLsN AwuMoy91zVFC7XjFiT/ZbjANeBu14qES1nDlk9AgBWpEhB2sYGpIEEzJVhBy7xTgVgWM W7HlmHlY8egcU36fBeOokW7cn407tF8snWv40TQxRvvskMI3DzMemAhmuQ7Tmt0ifwTL LgSC1noZfBOv/uOdI9qdP2fCLqTmZ/RvxHyDUsnkgQrGBaI7SEzDRV/1jqflWW468OQ3 9+dfol0OkJZjwCMU2g7aPY/jd5rlH6PXbuGkPzzfXpM4gbpe+hRXBleoSvx44D4UeymQ nHeg==; dara=google.com ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@linaro.org header.s=google header.b=vCNbjtn1; 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 pc2-20020a05620a840200b0078edc088d1asi28346738qkn.160.2024.04.30.12.44.08 for (version=TLS1_2 cipher=ECDHE-ECDSA-CHACHA20-POLY1305 bits=256/256); Tue, 30 Apr 2024 12:44:08 -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=vCNbjtn1; 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 1s1tNY-00073H-GX; Tue, 30 Apr 2024 15:43:04 -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 1s1tNW-00072D-Ra for qemu-devel@nongnu.org; Tue, 30 Apr 2024 15:43:02 -0400 Received: from mail-pf1-x42c.google.com ([2607:f8b0:4864:20::42c]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_128_GCM_SHA256:128) (Exim 4.90_1) (envelope-from ) id 1s1tNU-0006kQ-PG for qemu-devel@nongnu.org; Tue, 30 Apr 2024 15:43:02 -0400 Received: by mail-pf1-x42c.google.com with SMTP id d2e1a72fcca58-6f3f6aa1437so2738006b3a.3 for ; Tue, 30 Apr 2024 12:43:00 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; t=1714506178; x=1715110978; 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=Aco89HGhDUqxtzZSl7gQEXCIzX7CLhr3rgmAxnnD0GI=; b=vCNbjtn1tt5wRcjh52RhtlARq/neSk7+xPSuF4QqYIBUdWnlrIFn9FsNzj4CJ6Htsp YPN9mdL4I7FvP40watbnmVJrC6QdEgpO3bdQNwRKos2cdLpKY1mVOOHzGZVtEV3TB+so W4aMxPZ9CX55QWxiF8NMCj8vagkteWHgKtuxb+P9l7F+7fxfqSuYSMaDsFpm9lGtxfW+ pBAYJxZ8U0QHfllBkuI3tTATnSwbJalr27aUGp7M1TWvEN/lLdKuhRy2xk8V+sK/wAnM 17niRDjC9d+oVGX8uorhr/FofWmCr/u0cnlGU1IhwTjwbK/bMOHaohkNM/geWoSiQZEx tyzg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1714506178; x=1715110978; 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=Aco89HGhDUqxtzZSl7gQEXCIzX7CLhr3rgmAxnnD0GI=; b=gLJcYGbjZfVNX7wpRo5IAa6Zw2YUUnlJJHBdji2j4F86ojsgyrJI2Za0f2jM6ES7kn XyKIMD82vQrAcg+3/mKzftFlN7ouJTzK5To7vgOq6AZpcu8I9gyeLwJGSDitejxRsYf7 qdnjaStPx5rfZI4GDKoPVbnTShFnr+Q+e+fjpKSIOyb5Xe0VxWHzni16Mwjdr6U7UAzZ N6Z/yXU/7zylzZ0PsYmUBkRJ1nvnz7GfrOjGxFgR8vG724iFjJwB67NjsB09g9qUBqOa U9teLO715yauZ6XLZF01MlE8qF61T2QAQfPTdnow2E8kW0KWn/ekbHFVnnkmcqV0TcVj E/CA== X-Gm-Message-State: AOJu0YyfQLiaSCBprc9GkQcycuJkmg7hbVA1UYgrY+O53kPoKneCQnOo 6tnTfL1OAqf5Gq3uqesdqKvTkd5lExiF7gXWfSAYPldFeo+hwuCtWxtCvXKoBsGHvUsy/hp9Vvl z X-Received: by 2002:a05:6a21:3a83:b0:1af:66a9:d104 with SMTP id zv3-20020a056a213a8300b001af66a9d104mr451437pzb.1.1714506177934; Tue, 30 Apr 2024 12:42:57 -0700 (PDT) Received: from stoup.. (174-21-72-5.tukw.qwest.net. [174.21.72.5]) by smtp.gmail.com with ESMTPSA id d12-20020a63360c000000b005d880b41598sm20861523pga.94.2024.04.30.12.42.57 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 30 Apr 2024 12:42:57 -0700 (PDT) From: Richard Henderson To: qemu-devel@nongnu.org Cc: philmd@linaro.org, Alexander Monakov , Mikhail Romanov Subject: [PATCH v7 03/10] util/bufferiszero: Reorganize for early test for acceleration Date: Tue, 30 Apr 2024 12:42:46 -0700 Message-Id: <20240430194253.904768-4-richard.henderson@linaro.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240430194253.904768-1-richard.henderson@linaro.org> References: <20240430194253.904768-1-richard.henderson@linaro.org> MIME-Version: 1.0 Received-SPF: pass client-ip=2607:f8b0:4864:20::42c; envelope-from=richard.henderson@linaro.org; helo=mail-pf1-x42c.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 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 From: Alexander Monakov Test for length >= 256 inline, where is is often a constant. Before calling into the accelerated routine, sample three bytes from the buffer, which handles most non-zero buffers. Signed-off-by: Alexander Monakov Signed-off-by: Mikhail Romanov Message-Id: <20240206204809.9859-3-amonakov@ispras.ru> [rth: Use __builtin_constant_p; move the indirect call out of line.] Signed-off-by: Richard Henderson --- include/qemu/cutils.h | 32 ++++++++++++++++- util/bufferiszero.c | 84 +++++++++++++++++-------------------------- 2 files changed, 63 insertions(+), 53 deletions(-) diff --git a/include/qemu/cutils.h b/include/qemu/cutils.h index 92c927a6a3..741dade7cf 100644 --- a/include/qemu/cutils.h +++ b/include/qemu/cutils.h @@ -187,9 +187,39 @@ char *freq_to_str(uint64_t freq_hz); /* used to print char* safely */ #define STR_OR_NULL(str) ((str) ? (str) : "null") -bool buffer_is_zero(const void *buf, size_t len); +/* + * Check if a buffer is all zeroes. + */ + +bool buffer_is_zero_ool(const void *vbuf, size_t len); +bool buffer_is_zero_ge256(const void *vbuf, size_t len); bool test_buffer_is_zero_next_accel(void); +static inline bool buffer_is_zero_sample3(const char *buf, size_t len) +{ + /* + * For any reasonably sized buffer, these three samples come from + * three different cachelines. In qemu-img usage, we find that + * each byte eliminates more than half of all buffer testing. + * It is therefore critical to performance that the byte tests + * short-circuit, so that we do not pull in additional cache lines. + * Do not "optimize" this to !(a | b | c). + */ + return !buf[0] && !buf[len - 1] && !buf[len / 2]; +} + +#ifdef __OPTIMIZE__ +static inline bool buffer_is_zero(const void *buf, size_t len) +{ + return (__builtin_constant_p(len) && len >= 256 + ? buffer_is_zero_sample3(buf, len) && + buffer_is_zero_ge256(buf, len) + : buffer_is_zero_ool(buf, len)); +} +#else +#define buffer_is_zero buffer_is_zero_ool +#endif + /* * Implementation of ULEB128 (http://en.wikipedia.org/wiki/LEB128) * Input is limited to 14-bit numbers diff --git a/util/bufferiszero.c b/util/bufferiszero.c index 641d5f9b9e..972f394cbd 100644 --- a/util/bufferiszero.c +++ b/util/bufferiszero.c @@ -26,8 +26,9 @@ #include "qemu/bswap.h" #include "host/cpuinfo.h" -static bool -buffer_zero_int(const void *buf, size_t len) +static bool (*buffer_is_zero_accel)(const void *, size_t); + +static bool buffer_is_zero_integer(const void *buf, size_t len) { if (unlikely(len < 8)) { /* For a very small buffer, simply accumulate all the bytes. */ @@ -128,60 +129,38 @@ buffer_zero_avx2(const void *buf, size_t len) } #endif /* CONFIG_AVX2_OPT */ -/* - * Make sure that these variables are appropriately initialized when - * SSE2 is enabled on the compiler command-line, but the compiler is - * too old to support CONFIG_AVX2_OPT. - */ -#if defined(CONFIG_AVX2_OPT) -# define INIT_USED 0 -# define INIT_LENGTH 0 -# define INIT_ACCEL buffer_zero_int -#else -# ifndef __SSE2__ -# error "ISA selection confusion" -# endif -# define INIT_USED CPUINFO_SSE2 -# define INIT_LENGTH 64 -# define INIT_ACCEL buffer_zero_sse2 -#endif - -static unsigned used_accel = INIT_USED; -static unsigned length_to_accel = INIT_LENGTH; -static bool (*buffer_accel)(const void *, size_t) = INIT_ACCEL; - static unsigned __attribute__((noinline)) select_accel_cpuinfo(unsigned info) { /* Array is sorted in order of algorithm preference. */ static const struct { unsigned bit; - unsigned len; bool (*fn)(const void *, size_t); } all[] = { #ifdef CONFIG_AVX2_OPT - { CPUINFO_AVX2, 128, buffer_zero_avx2 }, + { CPUINFO_AVX2, buffer_zero_avx2 }, #endif - { CPUINFO_SSE2, 64, buffer_zero_sse2 }, - { CPUINFO_ALWAYS, 0, buffer_zero_int }, + { CPUINFO_SSE2, buffer_zero_sse2 }, + { CPUINFO_ALWAYS, buffer_is_zero_integer }, }; for (unsigned i = 0; i < ARRAY_SIZE(all); ++i) { if (info & all[i].bit) { - length_to_accel = all[i].len; - buffer_accel = all[i].fn; + buffer_is_zero_accel = all[i].fn; return all[i].bit; } } return 0; } -#if defined(CONFIG_AVX2_OPT) +static unsigned used_accel; + static void __attribute__((constructor)) init_accel(void) { used_accel = select_accel_cpuinfo(cpuinfo_init()); } -#endif /* CONFIG_AVX2_OPT */ + +#define INIT_ACCEL NULL bool test_buffer_is_zero_next_accel(void) { @@ -194,36 +173,37 @@ bool test_buffer_is_zero_next_accel(void) used_accel |= used; return used; } - -static bool select_accel_fn(const void *buf, size_t len) -{ - if (likely(len >= length_to_accel)) { - return buffer_accel(buf, len); - } - return buffer_zero_int(buf, len); -} - #else -#define select_accel_fn buffer_zero_int bool test_buffer_is_zero_next_accel(void) { return false; } + +#define INIT_ACCEL buffer_is_zero_integer #endif -/* - * Checks if a buffer is all zeroes - */ -bool buffer_is_zero(const void *buf, size_t len) +static bool (*buffer_is_zero_accel)(const void *, size_t) = INIT_ACCEL; + +bool buffer_is_zero_ool(const void *buf, size_t len) { if (unlikely(len == 0)) { return true; } + if (!buffer_is_zero_sample3(buf, len)) { + return false; + } + /* All bytes are covered for any len <= 3. */ + if (unlikely(len <= 3)) { + return true; + } - /* Fetch the beginning of the buffer while we select the accelerator. */ - __builtin_prefetch(buf); - - /* Use an optimized zero check if possible. Note that this also - includes a check for an unrolled loop over 64-bit integers. */ - return select_accel_fn(buf, len); + if (likely(len >= 256)) { + return buffer_is_zero_accel(buf, len); + } + return buffer_is_zero_integer(buf, len); +} + +bool buffer_is_zero_ge256(const void *buf, size_t len) +{ + return buffer_is_zero_accel(buf, len); }