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 604D24A56A for ; Sun, 2 Jun 2024 10:24:49 +0000 (UTC) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 883E268D5D8; Sun, 2 Jun 2024 13:24:46 +0300 (EEST) Received: from ursule.remlab.net (vps-a2bccee9.vps.ovh.net [51.75.19.47]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 3D6C268D3EB for ; Sun, 2 Jun 2024 13:24:39 +0300 (EEST) Received: from basile.remlab.net (localhost [IPv6:::1]) by ursule.remlab.net (Postfix) with ESMTP id B080BC0137 for ; Sun, 2 Jun 2024 13:24:38 +0300 (EEST) From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= To: ffmpeg-devel@ffmpeg.org Date: Sun, 2 Jun 2024 13:24:36 +0300 Message-ID: <20240602102438.12386-1-remi@remlab.net> X-Mailer: git-send-email 2.45.1 In-Reply-To: <20240601195613.44669-1-remi@remlab.net> References: <20240601195613.44669-1-remi@remlab.net> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH 3/5] lavc/vp8dsp: rework R-V V idct_dc_add4y 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-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: DCT-related FFmpeg functions often add an unsigned 8-bit sample to a signed 16-bit coefficient, then clip the result back to an unsigned 8-bit value. RISC-V has no signed 16-bit to unsigned 8-bit clip, so instead our most common sequence is: VWADDU.WV set SEW to 16 bits VMAX.VV zero # clip negative values to 0 set SEW to 8 bits VNCLIPU.WI # clip values over 255 to 255 and narrow Here we use a different sequence which does not require toggling the vector type. This assumes that the wide addend vector is biased by -128: VWADDU.WV VNCLIP.WI # clip values to signed 8-bit and narrow VXOR.VX 0x80 # flip sign bit (convert signed to unsigned) Also the VMAX is effectively replaced by a VXOR of half-width. In this function, this comes for free as we anyway add a constant to the wide vector in the prologue. On C908, this has no observable effects. On X60, this improves microbenchmarks by about 20%. --- libavcodec/riscv/vp7dsp_rvv.S | 2 +- libavcodec/riscv/vp8dsp_rvv.S | 14 +++++++------- 2 files changed, 8 insertions(+), 8 deletions(-) diff --git a/libavcodec/riscv/vp7dsp_rvv.S b/libavcodec/riscv/vp7dsp_rvv.S index 39b23c2e79..09dcbf3857 100644 --- a/libavcodec/riscv/vp7dsp_rvv.S +++ b/libavcodec/riscv/vp7dsp_rvv.S @@ -134,7 +134,7 @@ func ff_vp7_idct_dc_add4y_rvv, zve32x li t1, 23170 vlse16.v v8, (a1), t0 # block[0..3][0] vwmul.vx v0, v8, t1 - li t2, 0x20000 + li t2, 0x20000 - (128 << 18) vsetvli zero, zero, e32, m1, ta, ma vsra.vi v0, v0, 14 vmul.vx v0, v0, t1 diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S index 8ea0a0c9bd..458eebb306 100644 --- a/libavcodec/riscv/vp8dsp_rvv.S +++ b/libavcodec/riscv/vp8dsp_rvv.S @@ -125,31 +125,31 @@ endfunc func ff_vp8_idct_dc_add4y_rvv, zve32x li t0, 32 vsetivli zero, 4, e16, mf2, ta, ma + li t1, 4 - (128 << 3) vlse16.v v8, (a1), t0 - vadd.vi v8, v8, 4 + vadd.vx v8, v8, t1 vsra.vi v8, v8, 3 # fall through endfunc .variant_cc ff_vp78_idct_dc_add4y_rvv -# v8 = [dc0, dc1, dc2, dc3] +# v8 = [dc0 - 128, dc1 - 128, dc2 - 128, dc3 - 128] func ff_vp78_idct_dc_add4y_rvv, zve32x vsetivli zero, 16, e16, m2, ta, ma vid.v v4 + li a4, 4 vsrl.vi v4, v4, 2 + li t1, 128 vrgather.vv v0, v8, v4 # replicate each DC four times vsetvli zero, zero, e8, m1, ta, ma - li a4, 4 1: vle8.v v8, (a0) addi a4, a4, -1 vwaddu.wv v16, v0, v8 sh zero, (a1) - vsetvli zero, zero, e16, m2, ta, ma - vmax.vx v16, v16, zero + vnclip.wi v8, v16, 0 addi a1, a1, 32 - vsetvli zero, zero, e8, m1, ta, ma - vnclipu.wi v8, v16, 0 + vxor.vx v8, v8, t1 vse8.v v8, (a0) add a0, a0, a2 bnez a4, 1b -- 2.45.1 _______________________________________________ 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".