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 31A4447B47 for ; Mon, 2 Oct 2023 22:27:51 +0000 (UTC) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id CEBC968CB2C; Tue, 3 Oct 2023 01:27:49 +0300 (EEST) Received: from haasn.dev (haasn.dev [78.46.187.166]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id B1EA168CABE for ; Tue, 3 Oct 2023 01:27:42 +0300 (EEST) Received: from haasn.dev (unknown [10.30.0.2]) by haasn.dev (Postfix) with ESMTP id 6D95E4BC4E; Tue, 3 Oct 2023 00:27:41 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=simple/simple; d=haasn.xyz; s=mail; t=1696285661; bh=SOmV/wRqUnKauLMVUrjrHMJrkX7f6Zsp96eLA4fmHqU=; h=Date:From:To:Cc:Subject:In-Reply-To:References:From; b=XwDsFIttHcdfD1hT4sobvLMv9ImCpQkMAoKTH73tHIqeykSHGaqs6gFtdId32LhRd S8nZT/XkE3pq+7yhq62uMIP58mmUyIUbZbgHN2cPjrRMaYR3YOW9PyVPAA9tBBQuHj f4zvef/9qiZ3l1/UGLc82V+NShpCTTvvKO5tVoao= Date: Tue, 3 Oct 2023 00:27:41 +0200 Message-ID: <20231003002741.GB66943@haasn.xyz> From: Niklas Haas To: ffmpeg-devel@ffmpeg.org In-Reply-To: <20230928210848.95565-1-ffmpeg@haasn.xyz> References: <20230928210848.95565-1-ffmpeg@haasn.xyz> MIME-Version: 1.0 Content-Disposition: inline Subject: Re: [FFmpeg-devel] [PATCH] lavc/h274: transpose IDCT 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 Cc: Niklas Haas Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Archived-At: List-Archive: List-Post: On Thu, 28 Sep 2023 23:08:48 +0200 Niklas Haas wrote: > From: Niklas Haas > > This is mathematically equivalent to what we were doing before, but > gives subtly different results due to rounding (rows first vs columns > first). Doing it this way makes our film grain database generation match > reference implementation and now produces bit-exact outputs in my > testing. > > Rename the transposed variables to be a bit less confusing. > --- > libavcodec/h274.c | 20 ++++++++++---------- > 1 file changed, 10 insertions(+), 10 deletions(-) > > diff --git a/libavcodec/h274.c b/libavcodec/h274.c > index a5caf09564d..5709200322e 100644 > --- a/libavcodec/h274.c > +++ b/libavcodec/h274.c > @@ -59,13 +59,13 @@ static void init_slice_c(int8_t out[64][64], uint8_t h, uint8_t v, > // > // Note: To make the subsequent matrix multiplication cache friendlier, we > // store each *column* of the starting image in a *row* of `out` > - for (int y = 0; y <= freq_v; y++) { > - for (int x = 0; x <= freq_h; x += 4) { > + for (int l = 0; l <= freq_v; l++) { > + for (int k = 0; k <= freq_h; k += 4) { > uint16_t offset = seed % 2048; > - out[x + 0][y] = Gaussian_LUT[offset + 0]; > - out[x + 1][y] = Gaussian_LUT[offset + 1]; > - out[x + 2][y] = Gaussian_LUT[offset + 2]; > - out[x + 3][y] = Gaussian_LUT[offset + 3]; > + out[l][k + 0] = Gaussian_LUT[offset + 0]; > + out[l][k + 1] = Gaussian_LUT[offset + 1]; > + out[l][k + 2] = Gaussian_LUT[offset + 2]; > + out[l][k + 3] = Gaussian_LUT[offset + 3]; > prng_shift(&seed); > } > } > @@ -74,9 +74,9 @@ static void init_slice_c(int8_t out[64][64], uint8_t h, uint8_t v, > > // 64x64 inverse integer transform > for (int y = 0; y < 64; y++) { > - for (int x = 0; x <= freq_h; x++) { > + for (int x = 0; x <= freq_v; x++) { > int32_t sum = 0; > - for (int p = 0; p <= freq_v; p++) > + for (int p = 0; p <= freq_h; p++) > sum += R64T[y][p] * out[x][p]; > tmp[y][x] = (sum + 128) >> 8; > } > @@ -85,8 +85,8 @@ static void init_slice_c(int8_t out[64][64], uint8_t h, uint8_t v, > for (int y = 0; y < 64; y++) { > for (int x = 0; x < 64; x++) { > int32_t sum = 0; > - for (int p = 0; p <= freq_h; p++) > - sum += tmp[y][p] * R64T[x][p]; // R64T^T = R64 > + for (int p = 0; p <= freq_v; p++) > + sum += tmp[x][p] * R64T[y][p]; // R64T^T = R64 > // Renormalize and clip to [-127, 127] > out[y][x] = av_clip((sum + 128) >> 8, -127, 127); > } > -- > 2.42.0 > Merged as 22530ad1ce _______________________________________________ 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".