From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by master.gitmailbox.com (Postfix) with ESMTP id 4DDD6429D9 for ; Thu, 11 Aug 2022 00:15:08 +0000 (UTC) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 7DDBF68B8CC; Thu, 11 Aug 2022 03:15:06 +0300 (EEST) Received: from mail-ua1-f48.google.com (mail-ua1-f48.google.com [209.85.222.48]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 3643668B7C4 for ; Thu, 11 Aug 2022 03:15:00 +0300 (EEST) Received: by mail-ua1-f48.google.com with SMTP id f10so6432848uap.2 for ; Wed, 10 Aug 2022 17:15:00 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=content-transfer-encoding:in-reply-to:from:references:to :content-language:subject:user-agent:mime-version:date:message-id :from:to:cc; bh=xhefhxm2reyF39u0lmngABD7eI5yngI0KLiqRe6TyJM=; b=mQoxF8xw1ZqdlLMXZSOF0xlW3WfmSeVVElkkqbaDpWWaUljj+xpiZ2NVLF61lFaRVt OeEyYwIoKp1dvPdRKDVjSNyJbFDATGFXDzHKhEGRRjdXiPRBvp7O1DQbRNTtyqgeM/C7 JwJbFwRO0D2ah5aSw8fseYpI6fpe0+dxSojEaNxinTTCA80O1cGtd/wEwC8+GFWrhJ8y U5iekyGV6653Y6wJZ540YMwPHIZc7BCLhYUTmf2rcdFTJLe7nfxlnd1G3wELOAVzWr5Z 08nEgFrq0WhGMHO4hJMHdkr+ryIHg8ilHRtziMFerHYIue3vZRgskeUlJHdfSCYJViSV UjAQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=content-transfer-encoding:in-reply-to:from:references:to :content-language:subject:user-agent:mime-version:date:message-id :x-gm-message-state:from:to:cc; bh=xhefhxm2reyF39u0lmngABD7eI5yngI0KLiqRe6TyJM=; b=RvAnZlPKsRQ/kLDP969aXCwlPAs3KAgR43A/npXoYFHfcOZbii5Y4QelcAnosjVb59 3nqWJRRMx2RYmLSLCAg3XJOWeeb/CC1XllvtZJZ26lZUmiRwlzhxD1FiqLS3Ai2mWJlq Q9+MW7CQyiKMmerKZhV21a3UN/Zn6TsgfWqW97VIX5CtQAPcMFvBPWVd5ahI3eNY7W+C KyUjZBKQTEOH7Ml53QqMd+9rSR130SgtGPfAuroKs5WpsDusQnI5WLgOM4NVqSmG0Xq7 cskETN16kDiqwnvSZUQKpnlhw4CwdVROOlnYSJxpWTM5YE1v3sc0TuosxenFvBrE0VKd jT0Q== X-Gm-Message-State: ACgBeo1bkgfopoFpAHhYy4ciHpQVMNS9BMhM9BxtQL/aP3bI51J31m50 /RrQAvtHQLJJFQVbnD+jJmzk2wopBVI= X-Google-Smtp-Source: AA6agR6PANDsayJT0y5qI8dmDmfW3TW6hIUXKNGYrQn5IE0XwIcjYxvtC9QjNwEgk/5YxsNvznBgMA== X-Received: by 2002:ab0:2b06:0:b0:384:c4af:107c with SMTP id e6-20020ab02b06000000b00384c4af107cmr12661031uar.77.1660176897818; Wed, 10 Aug 2022 17:14:57 -0700 (PDT) Received: from [192.168.0.11] ([186.136.131.204]) by smtp.gmail.com with ESMTPSA id t24-20020a67c798000000b003884503a077sm726498vsk.25.2022.08.10.17.14.56 for (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Wed, 10 Aug 2022 17:14:57 -0700 (PDT) Message-ID: <314f36cb-5fe1-551b-81bc-b3b902dd6c79@gmail.com> Date: Wed, 10 Aug 2022 21:14:56 -0300 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:91.0) Gecko/20100101 Thunderbird/91.12.0 Content-Language: en-US To: ffmpeg-devel@ffmpeg.org References: <20220810204712.3123-9-timo@rothenpieler.org> <20220810225154.8435-1-timo@rothenpieler.org> From: James Almer In-Reply-To: <20220810225154.8435-1-timo@rothenpieler.org> Subject: Re: [FFmpeg-devel] [PATCH v2 09/11] avutil/half2float: use native _Float16 if available X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="us-ascii"; Format="flowed" Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Archived-At: List-Archive: List-Post: On 8/10/2022 7:51 PM, Timo Rothenpieler wrote: > _Float16 support was available on arm/aarch64 for a while, and with gcc > 12 was enabled on x86 as long as SSE2 is supported. > > If the target arch supports f16c, gcc emits fairly efficient assembly, > taking advantage of it. This is the case on x86-64-v3 or higher. > Same goes on arm, which has native float16 support. > On x86, without f16c, it emulates it in software using sse2 instructions. > > This has shown to perform rather poorly: > > _Float16 full SSE2 emulation: > frame=50074 fps=848 q=-0.0 size=N/A time=00:33:22.96 bitrate=N/A speed=33.9x > > _Float16 f16c accelerated (Zen2, --cpu=znver2): > frame=50636 fps=1965 q=-0.0 Lsize=N/A time=00:33:45.40 bitrate=N/A speed=78.6x > > classic half2float full software implementation: > frame=49926 fps=1605 q=-0.0 Lsize=N/A time=00:33:17.00 bitrate=N/A speed=64.2x > > Hence an additional check was introduced, that only enables use of > _Float16 on x86 if f16c is being utilized. > > On aarch64, a similar uplift in performance is seen: > > RPi4 half2float full software implementation: > frame= 6088 fps=126 q=-0.0 Lsize=N/A time=00:04:03.48 bitrate=N/A speed=5.06x > > RPi4 _Float16: > frame= 6103 fps=158 q=-0.0 Lsize=N/A time=00:04:04.08 bitrate=N/A speed=6.32x > > Since arm/aarch64 always natively support 16 bit floats, it can always > be considered fast there. > > I'm not aware of any additional platforms that currently support > _Float16. And if there are, they should be considered non-fast until > proven fast. > --- > configure | 13 +++++++++++++ > libavutil/float2half.c | 2 ++ > libavutil/float2half.h | 16 ++++++++++++++++ > libavutil/half2float.c | 4 ++++ > libavutil/half2float.h | 16 ++++++++++++++++ > 5 files changed, 51 insertions(+) > > diff --git a/configure b/configure > index 6761d0cb32..6ede9a5a8f 100755 > --- a/configure > +++ b/configure > @@ -2143,6 +2143,8 @@ ARCH_FEATURES=" > fast_64bit > fast_clz > fast_cmov > + fast_float16 > + float16 If HAVE_FLOAT16 is not going to be used, then don't export it here. Leave it as a configure internal variable. > local_aligned > simd_align_16 > simd_align_32 > @@ -5125,6 +5127,8 @@ elif enabled arm; then > ;; > esac > > + test_cflags -mfp16-format=ieee && add_cflags -mfp16-format=ieee > + > elif enabled avr32; then > > case $cpu in > @@ -6229,6 +6233,15 @@ check_builtin sync_val_compare_and_swap "" "int *ptr; int oldval, newval; __sync > check_builtin gmtime_r time.h "time_t *time; struct tm *tm; gmtime_r(time, tm)" > check_builtin localtime_r time.h "time_t *time; struct tm *tm; localtime_r(time, tm)" > > +check_builtin float16 "" "_Float16 f16var" > +if enabled float16; then > + if enabled x86; then > + test_cpp_condition stddef.h "defined(__F16C__)" && enable fast_float16 > + elif enabled arm || enabled aarch64; then > + enable fast_float16 > + fi > +fi > + > case "$custom_allocator" in > jemalloc) > # jemalloc by default does not use a prefix > diff --git a/libavutil/float2half.c b/libavutil/float2half.c > index dba14cef5d..7002612194 100644 > --- a/libavutil/float2half.c > +++ b/libavutil/float2half.c > @@ -20,6 +20,7 @@ > > void ff_init_float2half_tables(float2half_tables *t) > { > +#if !HAVE_FAST_FLOAT16 > for (int i = 0; i < 256; i++) { > int e = i - 127; > > @@ -50,4 +51,5 @@ void ff_init_float2half_tables(float2half_tables *t) > t->shifttable[i|0x100] = 13; > } > } > +#endif > } > diff --git a/libavutil/float2half.h b/libavutil/float2half.h > index b8c9cdfc4f..437666966b 100644 > --- a/libavutil/float2half.h > +++ b/libavutil/float2half.h > @@ -20,21 +20,37 @@ > #define AVUTIL_FLOAT2HALF_H > > #include > +#include "intfloat.h" > + > +#include "config.h" > > typedef struct float2half_tables { > +#if HAVE_FAST_FLOAT16 > + uint8_t dummy; > +#else > uint16_t basetable[512]; > uint8_t shifttable[512]; > +#endif > } float2half_tables; > > void ff_init_float2half_tables(float2half_tables *t); > > static inline uint16_t float2half(uint32_t f, const float2half_tables *t) > { > +#if HAVE_FAST_FLOAT16 > + union { > + _Float16 f; > + uint16_t i; > + } u; > + u.f = av_int2float(f); > + return u.i; > +#else > uint16_t h; > > h = t->basetable[(f >> 23) & 0x1ff] + ((f & 0x007fffff) >> t->shifttable[(f >> 23) & 0x1ff]); > > return h; > +#endif > } > > #endif /* AVUTIL_FLOAT2HALF_H */ > diff --git a/libavutil/half2float.c b/libavutil/half2float.c > index baac8e4093..ff198a8187 100644 > --- a/libavutil/half2float.c > +++ b/libavutil/half2float.c > @@ -18,6 +18,7 @@ > > #include "libavutil/half2float.h" > > +#if !HAVE_FAST_FLOAT16 > static uint32_t convertmantissa(uint32_t i) > { > int32_t m = i << 13; // Zero pad mantissa bits > @@ -33,9 +34,11 @@ static uint32_t convertmantissa(uint32_t i) > > return m | e; // Return combined number > } > +#endif > > void ff_init_half2float_tables(half2float_tables *t) > { > +#if !HAVE_FAST_FLOAT16 > t->mantissatable[0] = 0; > for (int i = 1; i < 1024; i++) > t->mantissatable[i] = convertmantissa(i); > @@ -60,4 +63,5 @@ void ff_init_half2float_tables(half2float_tables *t) > t->offsettable[31] = 2048; > t->offsettable[32] = 0; > t->offsettable[63] = 2048; > +#endif > } > diff --git a/libavutil/half2float.h b/libavutil/half2float.h > index cb58e44a1c..57ee8372fe 100644 > --- a/libavutil/half2float.h > +++ b/libavutil/half2float.h > @@ -20,22 +20,38 @@ > #define AVUTIL_HALF2FLOAT_H > > #include > +#include "intfloat.h" > + > +#include "config.h" > > typedef struct half2float_tables { > +#if HAVE_FAST_FLOAT16 > + uint8_t dummy; > +#else > uint32_t mantissatable[3072]; > uint32_t exponenttable[64]; > uint16_t offsettable[64]; > +#endif > } half2float_tables; > > void ff_init_half2float_tables(half2float_tables *t); > > static inline uint32_t half2float(uint16_t h, const half2float_tables *t) > { > +#if HAVE_FAST_FLOAT16 > + union { > + _Float16 f; > + uint16_t i; > + } u; > + u.i = h; > + return av_float2int(u.f); > +#else > uint32_t f; > > f = t->mantissatable[t->offsettable[h >> 10] + (h & 0x3ff)] + t->exponenttable[h >> 10]; > > return f; > +#endif > } > > #endif /* AVUTIL_HALF2FLOAT_H */ _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".