From patchwork Wed Mar 20 12:37:16 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Lauri Kasanen X-Patchwork-Id: 12367 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id A50584484A0 for ; Wed, 20 Mar 2019 14:39:46 +0200 (EET) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 844D668A90E; Wed, 20 Mar 2019 14:39:46 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mout.gmx.net (mout.gmx.net [212.227.17.22]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id A15D2689F00 for ; Wed, 20 Mar 2019 14:39:39 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=gmx.net; s=badeba3b8450; t=1553085579; bh=g5+ULeHXEmz4bLJ+XCFy15KFOzDT6PVwFw8rtzb3/wQ=; h=X-UI-Sender-Class:Date:From:To:Subject; b=N/AKLKGoLtfHxj3PThugYi0N8S5jiFFLu2CpFwpo/JIWSjBsfBMESreUuq8FMKynp Alen3TYLditRTD1KXRKiF5HCBq2Sdekj6DP/Yl1lkiaD/XOMuV3YTTS+HCwMJlF3hP i8kD8d9HcYxqiafdd/fLhqZagXPETAGp5WrPFLzA= X-UI-Sender-Class: 01bb95c1-4bf8-414a-932a-4f6e2808ef9c Received: from Valinor ([84.250.81.169]) by mail.gmx.com (mrgmx101 [212.227.17.174]) with ESMTPSA (Nemesis) id 0MDm4o-1hDhAe0NzS-00H8Fb for ; Wed, 20 Mar 2019 13:34:30 +0100 Date: Wed, 20 Mar 2019 14:37:16 +0200 From: Lauri Kasanen To: ffmpeg-devel@ffmpeg.org Message-Id: <20190320143716.c33a8270b6307a9c0eb75eb8@gmx.com> X-Mailer: Sylpheed 3.5.0 (GTK+ 2.18.6; x86_64-unknown-linux-gnu) Mime-Version: 1.0 X-Provags-ID: V03:K1:Nc39iJHswruufPLOWNAyQqwKDxwe3LvJ78Zcl10fnoSWKSP0JsV +Wr5PD7IWZG6FOM89Q0N9Q1ESpAwUWMdmIqn9n3TMTrS8ZkYAlxidG4Jx3BFx+0EnqTNuhM 3Y1MI/Zxe1RPdLjPpxeSHSU7XDIz74WfNWLy4GWbMhTeqNytxKLxmOtZV+RqhsfIOXH7k2z oCBCHC2hIYqn8/mXlSrOg== X-Spam-Flag: NO X-UI-Out-Filterresults: notjunk:1; V03:K0:7iFQVdbtArY=:hoi9N5tuNE4tiL3DH4tLt3 r0f6qi9poT8WvQXEabezhh6/DO4z/fvzBaxHJ+oTERUNUIKnxhl9d+d98nFUfFI+TCYLnr8XU ITa7MSnURs4Sx7/UPfXqeiKxrq6mOlXFK4MxhMRSO++lncUUTcRlzytMpx1SdwVbqH8cgIwLD zPts6MyrWTTehQrpkNhc1HyE2tKlFsy6yX27q5gH4MTU9N+04lBBTMQWd5z+rw1+UlvO6Rgns AOor6LJKH3gxvuEWBNbwtHynkCNDK3noTx1bhFauthISVAhcBhzR8XN5jxUBWnOoe8V+vxTIf 2WeItFz1p/NKemj01clBGusBjfgH1EpwYPkfWOmMrZLUg+nENo/qsHoVhu5SGifXORGFJXxZ6 q9027tW3QWBgTGSLTC9MCR0J4ijUiDwVNlmsucw06tkkMii77XVNCqxB+zwXVMMzE6wvQiq9V 9MupAxRU8WDhFkIS5ux3tEnSsEF8jxbkeYFyOi4VHfieuLstzejPabQpgkjeSk+VLrv2fB0Qr 2z7uzDmuw09ZO3NbPOTcZcxUu4aR3PuQYe5/uDKMmBAUYFQfsiV/f7dqiRpwR4XT3kqVFDYbG 5rIhYAjHNP8Px/KsYPrHcACl8g/BFJfS44QMmF6C2zEHpuDNGrJ/ckvkT6D3mJMNeBuCKbHvk Q5UgpLP3GBf3Q19JIaLhkFSIT42B4gUR9fjG06iV6UNCB0P2uHuzDiUAFq5uXbllbRJv0gc7y 0C6HbCt5BmBlc/iibG1Eb5Qw3ylGQxKDWQDpdGfkol1tRiraTHQbrtvGUaZXREP3sawbrn/vK kg7+zka5YRJY/uhHV6zY4OAB6jbYFo1zt/AXuZFi3+i6/pOXo5BtLjPyNiuBMWpNGhe84XSZD sPCGCQxukq4U/HpewlA6b7mQgTW2NTIo2rkCQ5u354tQlkhZpOuproimHNCZ0d Subject: [FFmpeg-devel] [PATCH] swscale/ppc: VSX-optimize yuv2rgb_full X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 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 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" ./ffmpeg -f lavfi -i yuvtestsrc=duration=1:size=1200x1440 \ -s 1200x1440 -f null -vframes 100 -pix_fmt $i -nostats \ -cpuflags 0 -v error - This uses 32-bit mul, so POWER8 only. The following output formats get about 4.5x speedup: rgb24 39980 UNITS in yuv2packed1, 32768 runs, 0 skips 8774 UNITS in yuv2packed1, 32768 runs, 0 skips bgr24 40069 UNITS in yuv2packed1, 32768 runs, 0 skips 8772 UNITS in yuv2packed1, 32766 runs, 2 skips rgba 39759 UNITS in yuv2packed1, 32768 runs, 0 skips 8681 UNITS in yuv2packed1, 32767 runs, 1 skips bgra 39729 UNITS in yuv2packed1, 32768 runs, 0 skips 8696 UNITS in yuv2packed1, 32766 runs, 2 skips argb 39766 UNITS in yuv2packed1, 32768 runs, 0 skips 8672 UNITS in yuv2packed1, 32766 runs, 2 skips bgra 39784 UNITS in yuv2packed1, 32768 runs, 0 skips 8659 UNITS in yuv2packed1, 32767 runs, 1 skips Signed-off-by: Lauri Kasanen --- libswscale/ppc/swscale_vsx.c | 303 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 303 insertions(+) -- 2.6.2 diff --git a/libswscale/ppc/swscale_vsx.c b/libswscale/ppc/swscale_vsx.c index 01eb46c..f20c11e 100644 --- a/libswscale/ppc/swscale_vsx.c +++ b/libswscale/ppc/swscale_vsx.c @@ -422,6 +422,248 @@ yuv2NBPSX(16, BE, 1, 16, int32_t) yuv2NBPSX(16, LE, 0, 16, int32_t) #endif +static av_always_inline void +yuv2rgb_full_1_vsx_template(SwsContext *c, const int16_t *buf0, + const int16_t *ubuf[2], const int16_t *vbuf[2], + const int16_t *abuf0, uint8_t *dest, int dstW, + int uvalpha, int y, enum AVPixelFormat target, + int hasAlpha) +{ + const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0]; + const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1]; + vector int16_t vy, vu, vv, A = vec_splat_s16(0), tmp16; + vector int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32, tmp32_2; + vector int32_t R_l, R_r, G_l, G_r, B_l, B_r; + vector uint16_t rd16, gd16, bd16; + vector uint8_t rd, bd, gd, ad, out0, out1, tmp8; + const vector uint16_t zero16 = vec_splat_u16(0); + const vector int32_t y_offset = vec_splats(c->yuv2rgb_y_offset); + const vector int32_t y_coeff = vec_splats(c->yuv2rgb_y_coeff); + const vector int32_t y_add = vec_splats(1 << 21); + const vector int32_t v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff); + const vector int32_t v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff); + const vector int32_t u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff); + const vector int32_t u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff); + const vector int32_t rgbclip = vec_splats(1 << 30); + const vector int32_t zero32 = vec_splat_s32(0); + const vector uint32_t shift2 = vec_splat_u32(2); + const vector uint32_t shift22 = vec_splats(22U); + const vector uint16_t sub7 = vec_splats((uint16_t) (128 << 7)); + const vector uint16_t sub8 = vec_splats((uint16_t) (128 << 8)); + const vector int16_t mul4 = vec_splat_s16(4); + const vector int16_t mul8 = vec_splat_s16(8); + const vector int16_t add64 = vec_splat_s16(64); + const vector uint16_t shift7 = vec_splat_u16(7); + const vector int16_t max255 = vec_splat_s16(255); + int i; + + // Various permutations + const vector uint8_t perm3rg0 = (vector uint8_t) {0x0, 0x10, 0, + 0x1, 0x11, 0, + 0x2, 0x12, 0, + 0x3, 0x13, 0, + 0x4, 0x14, 0, + 0x5 }; + const vector uint8_t perm3rg1 = (vector uint8_t) { 0x15, 0, + 0x6, 0x16, 0, + 0x7, 0x17, 0 }; + const vector uint8_t perm3tb0 = (vector uint8_t) {0x0, 0x1, 0x10, + 0x3, 0x4, 0x11, + 0x6, 0x7, 0x12, + 0x9, 0xa, 0x13, + 0xc, 0xd, 0x14, + 0xf }; + const vector uint8_t perm3tb1 = (vector uint8_t) { 0x0, 0x15, + 0x2, 0x3, 0x16, + 0x5, 0x6, 0x17 }; + + for (i = 0; i < dstW; i += 8) { // The x86 asm also overwrites padding bytes. + vy = vec_ld(0, &buf0[i]); + vy32_l = vec_unpackh(vy); + vy32_r = vec_unpackl(vy); + vy32_l = vec_sl(vy32_l, shift2); + vy32_r = vec_sl(vy32_r, shift2); + + vu = vec_ld(0, &ubuf0[i]); + vv = vec_ld(0, &vbuf0[i]); + if (uvalpha < 2048) { + vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub7); + vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub7); + + tmp32 = vec_mule(vu, mul4); + tmp32_2 = vec_mulo(vu, mul4); + vu32_l = vec_mergeh(tmp32, tmp32_2); + vu32_r = vec_mergel(tmp32, tmp32_2); + tmp32 = vec_mule(vv, mul4); + tmp32_2 = vec_mulo(vv, mul4); + vv32_l = vec_mergeh(tmp32, tmp32_2); + vv32_r = vec_mergel(tmp32, tmp32_2); + } else { + tmp16 = vec_ld(0, &ubuf1[i]); + vu = vec_add(vu, tmp16); + vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub8); + tmp16 = vec_ld(0, &vbuf1[i]); + vv = vec_add(vv, tmp16); + vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub8); + + vu32_l = vec_mule(vu, mul8); + vu32_r = vec_mulo(vu, mul8); + vv32_l = vec_mule(vv, mul8); + vv32_r = vec_mulo(vv, mul8); + } + + if (hasAlpha) { + A = vec_ld(0, &abuf0[i]); + A = vec_add(A, add64); + A = vec_sr(A, shift7); + A = vec_max(A, max255); + ad = vec_packsu(A, (vector int16_t) zero16); + } else { + ad = vec_splats((uint8_t) 255); + } + + vy32_l = vec_sub(vy32_l, y_offset); + vy32_r = vec_sub(vy32_r, y_offset); + vy32_l = vec_mul(vy32_l, y_coeff); + vy32_r = vec_mul(vy32_r, y_coeff); + vy32_l = vec_add(vy32_l, y_add); + vy32_r = vec_add(vy32_r, y_add); + + R_l = vec_mul(vv32_l, v2r_coeff); + R_l = vec_add(R_l, vy32_l); + R_r = vec_mul(vv32_r, v2r_coeff); + R_r = vec_add(R_r, vy32_r); + G_l = vec_mul(vv32_l, v2g_coeff); + tmp32 = vec_mul(vu32_l, u2g_coeff); + G_l = vec_add(G_l, vy32_l); + G_l = vec_add(G_l, tmp32); + G_r = vec_mul(vv32_r, v2g_coeff); + tmp32 = vec_mul(vu32_r, u2g_coeff); + G_r = vec_add(G_r, vy32_r); + G_r = vec_add(G_r, tmp32); + + B_l = vec_mul(vu32_l, u2b_coeff); + B_l = vec_add(B_l, vy32_l); + B_r = vec_mul(vu32_r, u2b_coeff); + B_r = vec_add(B_r, vy32_r); + + R_l = vec_max(R_l, zero32); + R_r = vec_max(R_r, zero32); + G_l = vec_max(G_l, zero32); + G_r = vec_max(G_r, zero32); + B_l = vec_max(B_l, zero32); + B_r = vec_max(B_r, zero32); + + R_l = vec_min(R_l, rgbclip); + R_r = vec_min(R_r, rgbclip); + G_l = vec_min(G_l, rgbclip); + G_r = vec_min(G_r, rgbclip); + B_l = vec_min(B_l, rgbclip); + B_r = vec_min(B_r, rgbclip); + + R_l = vec_sr(R_l, shift22); + R_r = vec_sr(R_r, shift22); + G_l = vec_sr(G_l, shift22); + G_r = vec_sr(G_r, shift22); + B_l = vec_sr(B_l, shift22); + B_r = vec_sr(B_r, shift22); + + rd16 = vec_packsu(R_l, R_r); + gd16 = vec_packsu(G_l, G_r); + bd16 = vec_packsu(B_l, B_r); + rd = vec_packsu(rd16, zero16); + gd = vec_packsu(gd16, zero16); + bd = vec_packsu(bd16, zero16); + + switch(target) { + case AV_PIX_FMT_RGB24: + out0 = vec_perm(rd, gd, perm3rg0); + out0 = vec_perm(out0, bd, perm3tb0); + out1 = vec_perm(rd, gd, perm3rg1); + out1 = vec_perm(out1, bd, perm3tb1); + + vec_vsx_st(out0, 0, dest); + vec_vsx_st(out1, 16, dest); + + dest += 24; + break; + case AV_PIX_FMT_BGR24: + out0 = vec_perm(bd, gd, perm3rg0); + out0 = vec_perm(out0, rd, perm3tb0); + out1 = vec_perm(bd, gd, perm3rg1); + out1 = vec_perm(out1, rd, perm3tb1); + + vec_vsx_st(out0, 0, dest); + vec_vsx_st(out1, 16, dest); + + dest += 24; + break; + case AV_PIX_FMT_BGRA: + out0 = vec_mergeh(bd, gd); + out1 = vec_mergeh(rd, ad); + + tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 0, dest); + tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 16, dest); + + dest += 32; + break; + case AV_PIX_FMT_RGBA: + out0 = vec_mergeh(rd, gd); + out1 = vec_mergeh(bd, ad); + + tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 0, dest); + tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 16, dest); + + dest += 32; + break; + case AV_PIX_FMT_ARGB: + out0 = vec_mergeh(ad, rd); + out1 = vec_mergeh(gd, bd); + + tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 0, dest); + tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 16, dest); + + dest += 32; + break; + case AV_PIX_FMT_ABGR: + out0 = vec_mergeh(ad, bd); + out1 = vec_mergeh(gd, rd); + + tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 0, dest); + tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); + vec_vsx_st(tmp8, 16, dest); + + dest += 32; + break; + } + } +} + +#define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha) \ +static void name ## ext ## _1_vsx(SwsContext *c, const int16_t *buf0, \ + const int16_t *ubuf[2], const int16_t *vbuf[2], \ + const int16_t *abuf0, uint8_t *dest, int dstW, \ + int uvalpha, int y) \ +{ \ + name ## base ## _1_vsx_template(c, buf0, ubuf, vbuf, abuf0, dest, \ + dstW, uvalpha, y, fmt, hasAlpha); \ +} + +YUV2RGBWRAPPER(yuv2, rgb_full, bgrx32_full, AV_PIX_FMT_BGRA, 0) +YUV2RGBWRAPPER(yuv2, rgb_full, rgbx32_full, AV_PIX_FMT_RGBA, 0) +YUV2RGBWRAPPER(yuv2, rgb_full, xrgb32_full, AV_PIX_FMT_ARGB, 0) +YUV2RGBWRAPPER(yuv2, rgb_full, xbgr32_full, AV_PIX_FMT_ABGR, 0) + +YUV2RGBWRAPPER(yuv2, rgb_full, rgb24_full, AV_PIX_FMT_RGB24, 0) +YUV2RGBWRAPPER(yuv2, rgb_full, bgr24_full, AV_PIX_FMT_BGR24, 0) + #endif /* !HAVE_BIGENDIAN */ #endif /* HAVE_VSX */ @@ -480,5 +722,66 @@ av_cold void ff_sws_init_swscale_vsx(SwsContext *c) #endif /* !HAVE_BIGENDIAN */ } } + + if (c->flags & SWS_BITEXACT) + return; + +#if !HAVE_BIGENDIAN + if (c->flags & SWS_FULL_CHR_H_INT) { + switch (dstFormat) { + case AV_PIX_FMT_RGB24: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + c->yuv2packed1 = yuv2rgb24_full_1_vsx; + } +#endif /* HAVE_POWER8 */ + break; + case AV_PIX_FMT_BGR24: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + c->yuv2packed1 = yuv2bgr24_full_1_vsx; + } +#endif /* HAVE_POWER8 */ + break; + case AV_PIX_FMT_BGRA: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + if (!c->needAlpha) { + c->yuv2packed1 = yuv2bgrx32_full_1_vsx; + } + } +#endif /* HAVE_POWER8 */ + break; + case AV_PIX_FMT_RGBA: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + if (!c->needAlpha) { + c->yuv2packed1 = yuv2rgbx32_full_1_vsx; + } + } +#endif /* HAVE_POWER8 */ + break; + case AV_PIX_FMT_ARGB: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + if (!c->needAlpha) { + c->yuv2packed1 = yuv2xrgb32_full_1_vsx; + } + } +#endif /* HAVE_POWER8 */ + break; + case AV_PIX_FMT_ABGR: +#if HAVE_POWER8 + if (cpu_flags & AV_CPU_FLAG_POWER8) { + if (!c->needAlpha) { + c->yuv2packed1 = yuv2xbgr32_full_1_vsx; + } + } +#endif /* HAVE_POWER8 */ + break; + } + } +#endif /* !HAVE_BIGENDIAN */ + #endif /* HAVE_VSX */ }