diff options
| author | Frank Barchard <fbarchard@google.com> | 2022-01-31 11:49:55 -0800 |
|---|---|---|
| committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2022-01-31 20:05:55 +0000 |
| commit | 804980bbab748fd0e180cd6e7d9292ff49baf704 (patch) | |
| tree | 79348716636fcffddf9630ce9654ad899cde68d6 /source/row_lsx.cc | |
| parent | 2c6bfc02d5265c95df69190c785f5d00d13bb444 (diff) | |
| download | libyuv-804980bbab748fd0e180cd6e7d9292ff49baf704.tar.gz | |
DetilePlane and unittest for NEON
Bug: libyuv:915, b/215425056
Change-Id: Iccab1ed3f6d385f02895d44faa94d198ad79d693
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/3424820
Reviewed-by: Justin Green <greenjustin@google.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Diffstat (limited to 'source/row_lsx.cc')
| -rw-r--r-- | source/row_lsx.cc | 640 |
1 files changed, 328 insertions, 312 deletions
diff --git a/source/row_lsx.cc b/source/row_lsx.cc index 6fe93b57..a445e636 100644 --- a/source/row_lsx.cc +++ b/source/row_lsx.cc @@ -21,139 +21,138 @@ extern "C" { #endif // Fill YUV -> RGB conversion constants into vectors -#define YUVTORGB_SETUP(yuvconst, vr, ub, vg, ug, yg, yb) \ - { \ - ub = __lsx_vreplgr2vr_h(yuvconst->kUVToB[0]); \ - vr = __lsx_vreplgr2vr_h(yuvconst->kUVToR[1]); \ - ug = __lsx_vreplgr2vr_h(yuvconst->kUVToG[0]); \ - vg = __lsx_vreplgr2vr_h(yuvconst->kUVToG[1]); \ - yg = __lsx_vreplgr2vr_h(yuvconst->kYToRgb[0]); \ - yb = __lsx_vreplgr2vr_w(yuvconst->kYBiasToRgb[0]); \ +#define YUVTORGB_SETUP(yuvconst, vr, ub, vg, ug, yg, yb) \ + { \ + ub = __lsx_vreplgr2vr_h(yuvconst->kUVToB[0]); \ + vr = __lsx_vreplgr2vr_h(yuvconst->kUVToR[1]); \ + ug = __lsx_vreplgr2vr_h(yuvconst->kUVToG[0]); \ + vg = __lsx_vreplgr2vr_h(yuvconst->kUVToG[1]); \ + yg = __lsx_vreplgr2vr_h(yuvconst->kYToRgb[0]); \ + yb = __lsx_vreplgr2vr_w(yuvconst->kYBiasToRgb[0]); \ } // Convert 8 pixels of YUV420 to RGB. -#define YUVTORGB(in_y, in_vu, vrub, vgug, \ - yg, yb, out_b, out_g, out_r) \ - { \ - __m128i y_ev, y_od, u_l, v_l; \ - __m128i tmp0, tmp1, tmp2, tmp3; \ - \ - tmp0 = __lsx_vilvl_b(in_y, in_y); \ - y_ev = __lsx_vmulwev_w_hu_h(tmp0, yg); \ - y_od = __lsx_vmulwod_w_hu_h(tmp0, yg); \ - y_ev = __lsx_vsrai_w(y_ev, 16); \ - y_od = __lsx_vsrai_w(y_od, 16); \ - y_ev = __lsx_vadd_w(y_ev, yb); \ - y_od = __lsx_vadd_w(y_od, yb); \ - in_vu = __lsx_vilvl_b(zero, in_vu); \ - in_vu = __lsx_vsub_h(in_vu, const_80); \ - u_l = __lsx_vmulwev_w_h(in_vu, vrub); \ - v_l = __lsx_vmulwod_w_h(in_vu, vrub); \ - tmp0 = __lsx_vadd_w(y_ev, u_l); \ - tmp1 = __lsx_vadd_w(y_od, u_l); \ - tmp2 = __lsx_vadd_w(y_ev, v_l); \ - tmp3 = __lsx_vadd_w(y_od, v_l); \ - tmp0 = __lsx_vsrai_w(tmp0, 6); \ - tmp1 = __lsx_vsrai_w(tmp1, 6); \ - tmp2 = __lsx_vsrai_w(tmp2, 6); \ - tmp3 = __lsx_vsrai_w(tmp3, 6); \ - tmp0 = __lsx_vclip255_w(tmp0); \ - tmp1 = __lsx_vclip255_w(tmp1); \ - tmp2 = __lsx_vclip255_w(tmp2); \ - tmp3 = __lsx_vclip255_w(tmp3); \ - out_b = __lsx_vpackev_h(tmp1, tmp0); \ - out_r = __lsx_vpackev_h(tmp3, tmp2); \ - tmp0 = __lsx_vdp2_w_h(in_vu, vgug); \ - tmp1 = __lsx_vsub_w(y_ev, tmp0); \ - tmp2 = __lsx_vsub_w(y_od, tmp0); \ - tmp1 = __lsx_vsrai_w(tmp1, 6); \ - tmp2 = __lsx_vsrai_w(tmp2, 6); \ - tmp1 = __lsx_vclip255_w(tmp1); \ - tmp2 = __lsx_vclip255_w(tmp2); \ - out_g = __lsx_vpackev_h(tmp2, tmp1); \ +#define YUVTORGB(in_y, in_vu, vrub, vgug, yg, yb, out_b, out_g, out_r) \ + { \ + __m128i y_ev, y_od, u_l, v_l; \ + __m128i tmp0, tmp1, tmp2, tmp3; \ + \ + tmp0 = __lsx_vilvl_b(in_y, in_y); \ + y_ev = __lsx_vmulwev_w_hu_h(tmp0, yg); \ + y_od = __lsx_vmulwod_w_hu_h(tmp0, yg); \ + y_ev = __lsx_vsrai_w(y_ev, 16); \ + y_od = __lsx_vsrai_w(y_od, 16); \ + y_ev = __lsx_vadd_w(y_ev, yb); \ + y_od = __lsx_vadd_w(y_od, yb); \ + in_vu = __lsx_vilvl_b(zero, in_vu); \ + in_vu = __lsx_vsub_h(in_vu, const_80); \ + u_l = __lsx_vmulwev_w_h(in_vu, vrub); \ + v_l = __lsx_vmulwod_w_h(in_vu, vrub); \ + tmp0 = __lsx_vadd_w(y_ev, u_l); \ + tmp1 = __lsx_vadd_w(y_od, u_l); \ + tmp2 = __lsx_vadd_w(y_ev, v_l); \ + tmp3 = __lsx_vadd_w(y_od, v_l); \ + tmp0 = __lsx_vsrai_w(tmp0, 6); \ + tmp1 = __lsx_vsrai_w(tmp1, 6); \ + tmp2 = __lsx_vsrai_w(tmp2, 6); \ + tmp3 = __lsx_vsrai_w(tmp3, 6); \ + tmp0 = __lsx_vclip255_w(tmp0); \ + tmp1 = __lsx_vclip255_w(tmp1); \ + tmp2 = __lsx_vclip255_w(tmp2); \ + tmp3 = __lsx_vclip255_w(tmp3); \ + out_b = __lsx_vpackev_h(tmp1, tmp0); \ + out_r = __lsx_vpackev_h(tmp3, tmp2); \ + tmp0 = __lsx_vdp2_w_h(in_vu, vgug); \ + tmp1 = __lsx_vsub_w(y_ev, tmp0); \ + tmp2 = __lsx_vsub_w(y_od, tmp0); \ + tmp1 = __lsx_vsrai_w(tmp1, 6); \ + tmp2 = __lsx_vsrai_w(tmp2, 6); \ + tmp1 = __lsx_vclip255_w(tmp1); \ + tmp2 = __lsx_vclip255_w(tmp2); \ + out_g = __lsx_vpackev_h(tmp2, tmp1); \ } // Convert I444 pixels of YUV420 to RGB. -#define I444TORGB(in_yy, in_u, in_v, ub, vr, ugvg, \ - yg, yb, out_b, out_g, out_r) \ - { \ - __m128i y_ev, y_od, u_ev, v_ev, u_od, v_od; \ - __m128i tmp0, tmp1, tmp2, tmp3; \ - \ - y_ev = __lsx_vmulwev_w_hu_h(in_yy, yg); \ - y_od = __lsx_vmulwod_w_hu_h(in_yy, yg); \ - y_ev = __lsx_vsrai_w(y_ev, 16); \ - y_od = __lsx_vsrai_w(y_od, 16); \ - y_ev = __lsx_vadd_w(y_ev, yb); \ - y_od = __lsx_vadd_w(y_od, yb); \ - in_u = __lsx_vsub_h(in_u, const_80); \ - in_v = __lsx_vsub_h(in_v, const_80); \ - u_ev = __lsx_vmulwev_w_h(in_u, ub); \ - u_od = __lsx_vmulwod_w_h(in_u, ub); \ - v_ev = __lsx_vmulwev_w_h(in_v, vr); \ - v_od = __lsx_vmulwod_w_h(in_v, vr); \ - tmp0 = __lsx_vadd_w(y_ev, u_ev); \ - tmp1 = __lsx_vadd_w(y_od, u_od); \ - tmp2 = __lsx_vadd_w(y_ev, v_ev); \ - tmp3 = __lsx_vadd_w(y_od, v_od); \ - tmp0 = __lsx_vsrai_w(tmp0, 6); \ - tmp1 = __lsx_vsrai_w(tmp1, 6); \ - tmp2 = __lsx_vsrai_w(tmp2, 6); \ - tmp3 = __lsx_vsrai_w(tmp3, 6); \ - tmp0 = __lsx_vclip255_w(tmp0); \ - tmp1 = __lsx_vclip255_w(tmp1); \ - tmp2 = __lsx_vclip255_w(tmp2); \ - tmp3 = __lsx_vclip255_w(tmp3); \ - out_b = __lsx_vpackev_h(tmp1, tmp0); \ - out_r = __lsx_vpackev_h(tmp3, tmp2); \ - u_ev = __lsx_vpackev_h(in_u, in_v); \ - u_od = __lsx_vpackod_h(in_u, in_v); \ - v_ev = __lsx_vdp2_w_h(u_ev, ugvg); \ - v_od = __lsx_vdp2_w_h(u_od, ugvg); \ - tmp0 = __lsx_vsub_w(y_ev, v_ev); \ - tmp1 = __lsx_vsub_w(y_od, v_od); \ - tmp0 = __lsx_vsrai_w(tmp0, 6); \ - tmp1 = __lsx_vsrai_w(tmp1, 6); \ - tmp0 = __lsx_vclip255_w(tmp0); \ - tmp1 = __lsx_vclip255_w(tmp1); \ - out_g = __lsx_vpackev_h(tmp1, tmp0); \ +#define I444TORGB(in_yy, in_u, in_v, ub, vr, ugvg, yg, yb, out_b, out_g, \ + out_r) \ + { \ + __m128i y_ev, y_od, u_ev, v_ev, u_od, v_od; \ + __m128i tmp0, tmp1, tmp2, tmp3; \ + \ + y_ev = __lsx_vmulwev_w_hu_h(in_yy, yg); \ + y_od = __lsx_vmulwod_w_hu_h(in_yy, yg); \ + y_ev = __lsx_vsrai_w(y_ev, 16); \ + y_od = __lsx_vsrai_w(y_od, 16); \ + y_ev = __lsx_vadd_w(y_ev, yb); \ + y_od = __lsx_vadd_w(y_od, yb); \ + in_u = __lsx_vsub_h(in_u, const_80); \ + in_v = __lsx_vsub_h(in_v, const_80); \ + u_ev = __lsx_vmulwev_w_h(in_u, ub); \ + u_od = __lsx_vmulwod_w_h(in_u, ub); \ + v_ev = __lsx_vmulwev_w_h(in_v, vr); \ + v_od = __lsx_vmulwod_w_h(in_v, vr); \ + tmp0 = __lsx_vadd_w(y_ev, u_ev); \ + tmp1 = __lsx_vadd_w(y_od, u_od); \ + tmp2 = __lsx_vadd_w(y_ev, v_ev); \ + tmp3 = __lsx_vadd_w(y_od, v_od); \ + tmp0 = __lsx_vsrai_w(tmp0, 6); \ + tmp1 = __lsx_vsrai_w(tmp1, 6); \ + tmp2 = __lsx_vsrai_w(tmp2, 6); \ + tmp3 = __lsx_vsrai_w(tmp3, 6); \ + tmp0 = __lsx_vclip255_w(tmp0); \ + tmp1 = __lsx_vclip255_w(tmp1); \ + tmp2 = __lsx_vclip255_w(tmp2); \ + tmp3 = __lsx_vclip255_w(tmp3); \ + out_b = __lsx_vpackev_h(tmp1, tmp0); \ + out_r = __lsx_vpackev_h(tmp3, tmp2); \ + u_ev = __lsx_vpackev_h(in_u, in_v); \ + u_od = __lsx_vpackod_h(in_u, in_v); \ + v_ev = __lsx_vdp2_w_h(u_ev, ugvg); \ + v_od = __lsx_vdp2_w_h(u_od, ugvg); \ + tmp0 = __lsx_vsub_w(y_ev, v_ev); \ + tmp1 = __lsx_vsub_w(y_od, v_od); \ + tmp0 = __lsx_vsrai_w(tmp0, 6); \ + tmp1 = __lsx_vsrai_w(tmp1, 6); \ + tmp0 = __lsx_vclip255_w(tmp0); \ + tmp1 = __lsx_vclip255_w(tmp1); \ + out_g = __lsx_vpackev_h(tmp1, tmp0); \ } // Pack and Store 8 ARGB values. -#define STOREARGB(in_a, in_r, in_g, in_b, pdst_argb) \ - { \ - __m128i temp0, temp1; \ - __m128i dst0, dst1; \ - \ - temp0 = __lsx_vpackev_b(in_g, in_b); \ - temp1 = __lsx_vpackev_b(in_a, in_r); \ - dst0 = __lsx_vilvl_h(temp1, temp0); \ - dst1 = __lsx_vilvh_h(temp1, temp0); \ - __lsx_vst(dst0, pdst_argb, 0); \ - __lsx_vst(dst1, pdst_argb, 16); \ - pdst_argb += 32; \ +#define STOREARGB(in_a, in_r, in_g, in_b, pdst_argb) \ + { \ + __m128i temp0, temp1; \ + __m128i dst0, dst1; \ + \ + temp0 = __lsx_vpackev_b(in_g, in_b); \ + temp1 = __lsx_vpackev_b(in_a, in_r); \ + dst0 = __lsx_vilvl_h(temp1, temp0); \ + dst1 = __lsx_vilvh_h(temp1, temp0); \ + __lsx_vst(dst0, pdst_argb, 0); \ + __lsx_vst(dst1, pdst_argb, 16); \ + pdst_argb += 32; \ } -#define RGBTOUV(_tmpb, _tmpg, _tmpr, _nexb, _nexg, _nexr, _dst0) \ - { \ - __m128i _tmp0, _tmp1, _tmp2, _tmp3; \ - __m128i _reg0, _reg1; \ - _tmp0 = __lsx_vaddwev_h_bu(_tmpb, _nexb); \ - _tmp1 = __lsx_vaddwod_h_bu(_tmpb, _nexb); \ - _tmp2 = __lsx_vaddwev_h_bu(_tmpg, _nexg); \ - _tmp3 = __lsx_vaddwod_h_bu(_tmpg, _nexg); \ - _reg0 = __lsx_vaddwev_h_bu(_tmpr, _nexr); \ - _reg1 = __lsx_vaddwod_h_bu(_tmpr, _nexr); \ - _tmpb = __lsx_vavgr_hu(_tmp0, _tmp1); \ - _tmpg = __lsx_vavgr_hu(_tmp2, _tmp3); \ - _tmpr = __lsx_vavgr_hu(_reg0, _reg1); \ - _reg0 = __lsx_vmadd_h(const_8080, const_112, _tmpb); \ - _reg1 = __lsx_vmadd_h(const_8080, const_112, _tmpr); \ - _reg0 = __lsx_vmsub_h(_reg0, const_74, _tmpg); \ - _reg1 = __lsx_vmsub_h(_reg1, const_94, _tmpg); \ - _reg0 = __lsx_vmsub_h(_reg0, const_38, _tmpr); \ - _reg1 = __lsx_vmsub_h(_reg1, const_18, _tmpb); \ - _dst0 = __lsx_vsrlni_b_h(_reg1, _reg0, 8); \ +#define RGBTOUV(_tmpb, _tmpg, _tmpr, _nexb, _nexg, _nexr, _dst0) \ + { \ + __m128i _tmp0, _tmp1, _tmp2, _tmp3; \ + __m128i _reg0, _reg1; \ + _tmp0 = __lsx_vaddwev_h_bu(_tmpb, _nexb); \ + _tmp1 = __lsx_vaddwod_h_bu(_tmpb, _nexb); \ + _tmp2 = __lsx_vaddwev_h_bu(_tmpg, _nexg); \ + _tmp3 = __lsx_vaddwod_h_bu(_tmpg, _nexg); \ + _reg0 = __lsx_vaddwev_h_bu(_tmpr, _nexr); \ + _reg1 = __lsx_vaddwod_h_bu(_tmpr, _nexr); \ + _tmpb = __lsx_vavgr_hu(_tmp0, _tmp1); \ + _tmpg = __lsx_vavgr_hu(_tmp2, _tmp3); \ + _tmpr = __lsx_vavgr_hu(_reg0, _reg1); \ + _reg0 = __lsx_vmadd_h(const_8080, const_112, _tmpb); \ + _reg1 = __lsx_vmadd_h(const_8080, const_112, _tmpr); \ + _reg0 = __lsx_vmsub_h(_reg0, const_74, _tmpg); \ + _reg1 = __lsx_vmsub_h(_reg1, const_94, _tmpg); \ + _reg0 = __lsx_vmsub_h(_reg0, const_38, _tmpr); \ + _reg1 = __lsx_vmsub_h(_reg1, const_18, _tmpb); \ + _dst0 = __lsx_vsrlni_b_h(_reg1, _reg0, 8); \ } void ARGB4444ToARGBRow_LSX(const uint8_t* src_argb4444, @@ -177,8 +176,8 @@ void ARGB4444ToARGBRow_LSX(const uint8_t* src_argb4444, reg2 = __lsx_vslli_b(tmp2, 4); reg1 = __lsx_vsrli_b(tmp1, 4); reg3 = __lsx_vsrli_b(tmp3, 4); - DUP4_ARG2(__lsx_vor_v, tmp0, reg0, tmp1, reg1, tmp2, reg2, - tmp3, reg3, tmp0, tmp1, tmp2, tmp3); + DUP4_ARG2(__lsx_vor_v, tmp0, reg0, tmp1, reg1, tmp2, reg2, tmp3, reg3, tmp0, + tmp1, tmp2, tmp3); dst0 = __lsx_vilvl_b(tmp1, tmp0); dst2 = __lsx_vilvl_b(tmp3, tmp2); dst1 = __lsx_vilvh_b(tmp1, tmp0); @@ -352,9 +351,9 @@ void ARGB1555ToYRow_LSX(const uint8_t* src_argb1555, __m128i src0, src1; __m128i tmp0, tmp1, tmpb, tmpg, tmpr; __m128i reg0, reg1, reg2, dst0; - __m128i const_66 = __lsx_vldi(66); + __m128i const_66 = __lsx_vldi(66); __m128i const_129 = __lsx_vldi(129); - __m128i const_25 = __lsx_vldi(25); + __m128i const_25 = __lsx_vldi(25); __m128i const_1080 = {0x1080108010801080, 0x1080108010801080}; __m128i shuff = {0x0B030A0209010800, 0x0F070E060D050C04}; @@ -406,15 +405,15 @@ void ARGB1555ToUVRow_LSX(const uint8_t* src_argb1555, __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i reg0, reg1, reg2, reg3, dst0; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_argb1555, 0, src_argb1555, 16, - next_argb1555, 0, next_argb1555, 16, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_argb1555, 0, src_argb1555, 16, next_argb1555, 0, + next_argb1555, 16, src0, src1, src2, src3); DUP2_ARG2(__lsx_vpickev_b, src1, src0, src3, src2, tmp0, tmp2); DUP2_ARG2(__lsx_vpickod_b, src1, src0, src3, src2, tmp1, tmp3); tmpb = __lsx_vandi_b(tmp0, 0x1F); @@ -465,9 +464,9 @@ void RGB565ToYRow_LSX(const uint8_t* src_rgb565, uint8_t* dst_y, int width) { __m128i src0, src1; __m128i tmp0, tmp1, tmpb, tmpg, tmpr; __m128i reg0, reg1, dst0; - __m128i const_66 = __lsx_vldi(66); + __m128i const_66 = __lsx_vldi(66); __m128i const_129 = __lsx_vldi(129); - __m128i const_25 = __lsx_vldi(25); + __m128i const_25 = __lsx_vldi(25); __m128i const_1080 = {0x1080108010801080, 0x1080108010801080}; __m128i shuff = {0x0B030A0209010800, 0x0F070E060D050C04}; @@ -517,15 +516,15 @@ void RGB565ToUVRow_LSX(const uint8_t* src_rgb565, __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i reg0, reg1, reg2, reg3, dst0; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_rgb565, 0, src_rgb565, 16, - next_rgb565, 0, next_rgb565, 16, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_rgb565, 0, src_rgb565, 16, next_rgb565, 0, + next_rgb565, 16, src0, src1, src2, src3); DUP2_ARG2(__lsx_vpickev_b, src1, src0, src3, src2, tmp0, tmp2); DUP2_ARG2(__lsx_vpickod_b, src1, src0, src3, src2, tmp1, tmp3); tmpb = __lsx_vandi_b(tmp0, 0x1F); @@ -611,10 +610,10 @@ void RGB24ToUVRow_LSX(const uint8_t* src_rgb24, __m128i nex0, nex1, nex2, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; __m128i shuff0_b = {0x15120F0C09060300, 0x00000000001E1B18}; __m128i shuff1_b = {0x0706050403020100, 0x1D1A1714110A0908}; @@ -630,12 +629,18 @@ void RGB24ToUVRow_LSX(const uint8_t* src_rgb24, nex0 = __lsx_vld(next_rgb24, 0); nex1 = __lsx_vld(next_rgb24, 16); nex2 = __lsx_vld(next_rgb24, 32); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_b, nex1, nex0, shuff0_b, tmpb, nexb); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_g, nex1, nex0, shuff0_g, tmpg, nexg); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_r, nex1, nex0, shuff0_r, tmpr, nexr); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpb, shuff1_b, nex2, nexb, shuff1_b, tmpb, nexb); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpg, shuff1_g, nex2, nexg, shuff1_g, tmpg, nexg); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpr, shuff1_r, nex2, nexr, shuff1_r, tmpr, nexr); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_b, nex1, nex0, shuff0_b, tmpb, + nexb); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_g, nex1, nex0, shuff0_g, tmpg, + nexg); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_r, nex1, nex0, shuff0_r, tmpr, + nexr); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpb, shuff1_b, nex2, nexb, shuff1_b, tmpb, + nexb); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpg, shuff1_g, nex2, nexg, shuff1_g, tmpg, + nexg); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpr, shuff1_r, nex2, nexr, shuff1_r, tmpr, + nexr); RGBTOUV(tmpb, tmpg, tmpr, nexb, nexg, nexr, dst0); __lsx_vstelm_d(dst0, dst_u, 0, 0); __lsx_vstelm_d(dst0, dst_v, 0, 1); @@ -691,10 +696,10 @@ void RAWToUVRow_LSX(const uint8_t* src_raw, __m128i nex0, nex1, nex2, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; __m128i shuff0_r = {0x15120F0C09060300, 0x00000000001E1B18}; __m128i shuff1_r = {0x0706050403020100, 0x1D1A1714110A0908}; @@ -710,12 +715,18 @@ void RAWToUVRow_LSX(const uint8_t* src_raw, nex0 = __lsx_vld(next_raw, 0); nex1 = __lsx_vld(next_raw, 16); nex2 = __lsx_vld(next_raw, 32); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_b, nex1, nex0, shuff0_b, tmpb, nexb); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_g, nex1, nex0, shuff0_g, tmpg, nexg); - DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_r, nex1, nex0, shuff0_r, tmpr, nexr); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpb, shuff1_b, nex2, nexb, shuff1_b, tmpb, nexb); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpg, shuff1_g, nex2, nexg, shuff1_g, tmpg, nexg); - DUP2_ARG3(__lsx_vshuf_b, src2, tmpr, shuff1_r, nex2, nexr, shuff1_r, tmpr, nexr); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_b, nex1, nex0, shuff0_b, tmpb, + nexb); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_g, nex1, nex0, shuff0_g, tmpg, + nexg); + DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0_r, nex1, nex0, shuff0_r, tmpr, + nexr); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpb, shuff1_b, nex2, nexb, shuff1_b, tmpb, + nexb); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpg, shuff1_g, nex2, nexg, shuff1_g, tmpg, + nexg); + DUP2_ARG3(__lsx_vshuf_b, src2, tmpr, shuff1_r, nex2, nexr, shuff1_r, tmpr, + nexr); RGBTOUV(tmpb, tmpg, tmpr, nexb, nexg, nexr, dst0); __lsx_vstelm_d(dst0, dst_u, 0, 0); __lsx_vstelm_d(dst0, dst_v, 0, 1); @@ -739,19 +750,19 @@ void NV12ToARGBRow_LSX(const uint8_t* src_y, __m128i out_b, out_g, out_r; __m128i const_80 = __lsx_vldi(0x480); __m128i alpha = __lsx_vldi(0xFF); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); vec_vrub = __lsx_vilvl_h(vec_vr, vec_ub); vec_vgug = __lsx_vilvl_h(vec_vg, vec_ug); for (x = 0; x < len; x++) { - vec_y = __lsx_vld(src_y, 0); + vec_y = __lsx_vld(src_y, 0); vec_vu = __lsx_vld(src_uv, 0); - YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, - out_b, out_g, out_r); + YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, out_b, out_g, + out_r); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); - src_y += 8; + src_y += 8; src_uv += 8; } } @@ -768,17 +779,17 @@ void NV12ToRGB565Row_LSX(const uint8_t* src_y, __m128i vec_vrub, vec_vgug; __m128i out_b, out_g, out_r; __m128i const_80 = __lsx_vldi(0x480); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); vec_vrub = __lsx_vilvl_h(vec_vr, vec_ub); vec_vgug = __lsx_vilvl_h(vec_vg, vec_ug); for (x = 0; x < len; x++) { - vec_y = __lsx_vld(src_y, 0); + vec_y = __lsx_vld(src_y, 0); vec_vu = __lsx_vld(src_uv, 0); - YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, - out_b, out_g, out_r); + YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, out_b, out_g, + out_r); out_b = __lsx_vsrli_h(out_b, 3); out_g = __lsx_vsrli_h(out_g, 2); out_r = __lsx_vsrli_h(out_r, 3); @@ -787,7 +798,7 @@ void NV12ToRGB565Row_LSX(const uint8_t* src_y, out_r = __lsx_vor_v(out_r, out_g); out_r = __lsx_vor_v(out_r, out_b); __lsx_vst(out_r, dst_rgb565, 0); - src_y += 8; + src_y += 8; src_uv += 8; dst_rgb565 += 16; } @@ -806,19 +817,19 @@ void NV21ToARGBRow_LSX(const uint8_t* src_y, __m128i out_b, out_g, out_r; __m128i const_80 = __lsx_vldi(0x480); __m128i alpha = __lsx_vldi(0xFF); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); vec_ubvr = __lsx_vilvl_h(vec_ub, vec_vr); vec_ugvg = __lsx_vilvl_h(vec_ug, vec_vg); for (x = 0; x < len; x++) { - vec_y = __lsx_vld(src_y, 0); + vec_y = __lsx_vld(src_y, 0); vec_uv = __lsx_vld(src_vu, 0); - YUVTORGB(vec_y, vec_uv, vec_ubvr, vec_ugvg, vec_yg, vec_yb, - out_r, out_g, out_b); + YUVTORGB(vec_y, vec_uv, vec_ubvr, vec_ugvg, vec_yg, vec_yb, out_r, out_g, + out_b); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); - src_y += 8; + src_y += 8; src_vu += 8; } } @@ -831,7 +842,7 @@ void SobelRow_LSX(const uint8_t* src_sobelx, int len = width / 16; __m128i src0, src1, tmp0; __m128i out0, out1, out2, out3; - __m128i alpha = __lsx_vldi(0xFF); + __m128i alpha = __lsx_vldi(0xFF); __m128i shuff0 = {0x1001010110000000, 0x1003030310020202}; __m128i shuff1 = __lsx_vaddi_bu(shuff0, 0x04); __m128i shuff2 = __lsx_vaddi_bu(shuff1, 0x04); @@ -915,11 +926,11 @@ void ARGBToYJRow_LSX(const uint8_t* src_argb, uint8_t* dst_y, int width) { __m128i reg0, reg1; __m128i const_128 = __lsx_vldi(0x480); __m128i const_150 = __lsx_vldi(0x96); - __m128i const_br = {0x4D1D4D1D4D1D4D1D, 0x4D1D4D1D4D1D4D1D}; + __m128i const_br = {0x4D1D4D1D4D1D4D1D, 0x4D1D4D1D4D1D4D1D}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, - src_argb, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, src_argb, 48, + src0, src1, src2, src3); tmp0 = __lsx_vpickev_b(src1, src0); tmp1 = __lsx_vpickod_b(src1, src0); tmp2 = __lsx_vpickev_b(src3, src2); @@ -942,12 +953,12 @@ void BGRAToYRow_LSX(const uint8_t* src_bgra, uint8_t* dst_y, int width) { __m128i tmp0, tmp1, tmp2, tmp3; __m128i reg0, reg1; __m128i const_129 = __lsx_vldi(0x81); - __m128i const_br = {0x1942194219421942, 0x1942194219421942}; + __m128i const_br = {0x1942194219421942, 0x1942194219421942}; __m128i const_1080 = {0x1080108010801080, 0x1080108010801080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_bgra, 0, src_bgra, 16, src_bgra, 32, - src_bgra, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_bgra, 0, src_bgra, 16, src_bgra, 32, src_bgra, 48, + src0, src1, src2, src3); tmp0 = __lsx_vpickod_b(src1, src0); tmp1 = __lsx_vpickev_b(src1, src0); tmp2 = __lsx_vpickod_b(src3, src2); @@ -976,17 +987,17 @@ void BGRAToUVRow_LSX(const uint8_t* src_bgra, __m128i tmp0, tmp1, tmp2, tmp3, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_bgra, 0, src_bgra, 16, src_bgra, 32, - src_bgra, 48, src0, src1, src2, src3); - DUP4_ARG2(__lsx_vld, next_bgra, 0, next_bgra, 16, next_bgra, 32, - next_bgra, 48, nex0, nex1, nex2, nex3); + DUP4_ARG2(__lsx_vld, src_bgra, 0, src_bgra, 16, src_bgra, 32, src_bgra, 48, + src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, next_bgra, 0, next_bgra, 16, next_bgra, 32, next_bgra, + 48, nex0, nex1, nex2, nex3); tmp0 = __lsx_vpickod_b(src1, src0); tmp1 = __lsx_vpickev_b(src1, src0); tmp2 = __lsx_vpickod_b(src3, src2); @@ -1018,12 +1029,12 @@ void ABGRToYRow_LSX(const uint8_t* src_abgr, uint8_t* dst_y, int width) { __m128i tmp0, tmp1, tmp2, tmp3; __m128i reg0, reg1; __m128i const_129 = __lsx_vldi(0x81); - __m128i const_br = {0x1942194219421942, 0x1942194219421942}; + __m128i const_br = {0x1942194219421942, 0x1942194219421942}; __m128i const_1080 = {0x1080108010801080, 0x1080108010801080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_abgr, 0, src_abgr, 16, src_abgr, 32, - src_abgr, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_abgr, 0, src_abgr, 16, src_abgr, 32, src_abgr, 48, + src0, src1, src2, src3); tmp0 = __lsx_vpickev_b(src1, src0); tmp1 = __lsx_vpickod_b(src1, src0); tmp2 = __lsx_vpickev_b(src3, src2); @@ -1052,17 +1063,17 @@ void ABGRToUVRow_LSX(const uint8_t* src_abgr, __m128i tmp0, tmp1, tmp2, tmp3, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_abgr, 0, src_abgr, 16, src_abgr, 32, - src_abgr, 48, src0, src1, src2, src3); - DUP4_ARG2(__lsx_vld, next_abgr, 0, next_abgr, 16, next_abgr, 32, - next_abgr, 48, nex0, nex1, nex2, nex3); + DUP4_ARG2(__lsx_vld, src_abgr, 0, src_abgr, 16, src_abgr, 32, src_abgr, 48, + src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, next_abgr, 0, next_abgr, 16, next_abgr, 32, next_abgr, + 48, nex0, nex1, nex2, nex3); tmp0 = __lsx_vpickev_b(src1, src0); tmp1 = __lsx_vpickod_b(src1, src0); tmp2 = __lsx_vpickev_b(src3, src2); @@ -1094,12 +1105,12 @@ void RGBAToYRow_LSX(const uint8_t* src_rgba, uint8_t* dst_y, int width) { __m128i tmp0, tmp1, tmp2, tmp3; __m128i reg0, reg1; __m128i const_129 = __lsx_vldi(0x81); - __m128i const_br = {0x4219421942194219, 0x4219421942194219}; + __m128i const_br = {0x4219421942194219, 0x4219421942194219}; __m128i const_1080 = {0x1080108010801080, 0x1080108010801080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_rgba, 0, src_rgba, 16, src_rgba, 32, - src_rgba, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_rgba, 0, src_rgba, 16, src_rgba, 32, src_rgba, 48, + src0, src1, src2, src3); tmp0 = __lsx_vpickod_b(src1, src0); tmp1 = __lsx_vpickev_b(src1, src0); tmp2 = __lsx_vpickod_b(src3, src2); @@ -1128,17 +1139,17 @@ void RGBAToUVRow_LSX(const uint8_t* src_rgba, __m128i tmp0, tmp1, tmp2, tmp3, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; __m128i const_112 = __lsx_vldi(0x438); - __m128i const_74 = __lsx_vldi(0x425); - __m128i const_38 = __lsx_vldi(0x413); - __m128i const_94 = __lsx_vldi(0x42F); - __m128i const_18 = __lsx_vldi(0x409); + __m128i const_74 = __lsx_vldi(0x425); + __m128i const_38 = __lsx_vldi(0x413); + __m128i const_94 = __lsx_vldi(0x42F); + __m128i const_18 = __lsx_vldi(0x409); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_rgba, 0, src_rgba, 16, src_rgba, 32, - src_rgba, 48, src0, src1, src2, src3); - DUP4_ARG2(__lsx_vld, next_rgba, 0, next_rgba, 16, next_rgba, 32, - next_rgba, 48, nex0, nex1, nex2, nex3); + DUP4_ARG2(__lsx_vld, src_rgba, 0, src_rgba, 16, src_rgba, 32, src_rgba, 48, + src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, next_rgba, 0, next_rgba, 16, next_rgba, 32, next_rgba, + 48, nex0, nex1, nex2, nex3); tmp0 = __lsx_vpickod_b(src1, src0); tmp1 = __lsx_vpickev_b(src1, src0); tmp2 = __lsx_vpickod_b(src3, src2); @@ -1174,20 +1185,20 @@ void ARGBToUVJRow_LSX(const uint8_t* src_argb, __m128i src0, src1, src2, src3; __m128i nex0, nex1, nex2, nex3; __m128i tmp0, tmp1, tmp2, tmp3; - __m128i reg0, reg1, dst0; + __m128i reg0, reg1, dst0; __m128i tmpb, tmpg, tmpr, nexb, nexg, nexr; - __m128i const_63 = __lsx_vldi(0x43F); - __m128i const_42 = __lsx_vldi(0x42A); - __m128i const_21 = __lsx_vldi(0x415); - __m128i const_53 = __lsx_vldi(0x435); - __m128i const_10 = __lsx_vldi(0x40A); + __m128i const_63 = __lsx_vldi(0x43F); + __m128i const_42 = __lsx_vldi(0x42A); + __m128i const_21 = __lsx_vldi(0x415); + __m128i const_53 = __lsx_vldi(0x435); + __m128i const_10 = __lsx_vldi(0x40A); __m128i const_8080 = {0x8080808080808080, 0x8080808080808080}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, - src_argb, 48, src0, src1, src2, src3); - DUP4_ARG2(__lsx_vld, next_argb, 0, next_argb, 16, next_argb, 32, - next_argb, 48, nex0, nex1, nex2, nex3); + DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, src_argb, 48, + src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, next_argb, 0, next_argb, 16, next_argb, 32, next_argb, + 48, nex0, nex1, nex2, nex3); tmp0 = __lsx_vpickev_b(src1, src0); tmp1 = __lsx_vpickod_b(src1, src0); tmp2 = __lsx_vpickev_b(src3, src2); @@ -1240,26 +1251,26 @@ void I444ToARGBRow_LSX(const uint8_t* src_y, __m128i vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb, vec_ugvg; __m128i const_80 = __lsx_vldi(0x480); __m128i alpha = __lsx_vldi(0xFF); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); vec_ugvg = __lsx_vilvl_h(vec_ug, vec_vg); for (x = 0; x < len; x++) { - vec_y = __lsx_vld(src_y, 0); - vec_u = __lsx_vld(src_u, 0); - vec_v = __lsx_vld(src_v, 0); + vec_y = __lsx_vld(src_y, 0); + vec_u = __lsx_vld(src_u, 0); + vec_v = __lsx_vld(src_v, 0); vec_yl = __lsx_vilvl_b(vec_y, vec_y); vec_ul = __lsx_vilvl_b(zero, vec_u); vec_vl = __lsx_vilvl_b(zero, vec_v); - I444TORGB(vec_yl, vec_ul, vec_vl, vec_ub, vec_vr, vec_ugvg, - vec_yg, vec_yb, out_b, out_g, out_r); + I444TORGB(vec_yl, vec_ul, vec_vl, vec_ub, vec_vr, vec_ugvg, vec_yg, vec_yb, + out_b, out_g, out_r); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); vec_yh = __lsx_vilvh_b(vec_y, vec_y); vec_uh = __lsx_vilvh_b(zero, vec_u); vec_vh = __lsx_vilvh_b(zero, vec_v); - I444TORGB(vec_yh, vec_uh, vec_vh, vec_ub, vec_vr, vec_ugvg, - vec_yg, vec_yb, out_b, out_g, out_r); + I444TORGB(vec_yh, vec_uh, vec_vh, vec_ub, vec_vr, vec_ugvg, vec_yg, vec_yb, + out_b, out_g, out_r); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); src_y += 16; src_u += 16; @@ -1283,37 +1294,37 @@ void I400ToARGBRow_LSX(const uint8_t* src_y, for (x = 0; x < len; x++) { vec_y = __lsx_vld(src_y, 0); vec_yl = __lsx_vilvl_b(vec_y, vec_y); - y_ev = __lsx_vmulwev_w_hu_h(vec_yl, vec_yg); - y_od = __lsx_vmulwod_w_hu_h(vec_yl, vec_yg); - y_ev = __lsx_vsrai_w(y_ev, 16); - y_od = __lsx_vsrai_w(y_od, 16); - y_ev = __lsx_vadd_w(y_ev, vec_yb); - y_od = __lsx_vadd_w(y_od, vec_yb); - y_ev = __lsx_vsrai_w(y_ev, 6); - y_od = __lsx_vsrai_w(y_od, 6); - y_ev = __lsx_vclip255_w(y_ev); - y_od = __lsx_vclip255_w(y_od); - out0 = __lsx_vpackev_h(y_od, y_ev); + y_ev = __lsx_vmulwev_w_hu_h(vec_yl, vec_yg); + y_od = __lsx_vmulwod_w_hu_h(vec_yl, vec_yg); + y_ev = __lsx_vsrai_w(y_ev, 16); + y_od = __lsx_vsrai_w(y_od, 16); + y_ev = __lsx_vadd_w(y_ev, vec_yb); + y_od = __lsx_vadd_w(y_od, vec_yb); + y_ev = __lsx_vsrai_w(y_ev, 6); + y_od = __lsx_vsrai_w(y_od, 6); + y_ev = __lsx_vclip255_w(y_ev); + y_od = __lsx_vclip255_w(y_od); + out0 = __lsx_vpackev_h(y_od, y_ev); temp0 = __lsx_vpackev_b(out0, out0); temp1 = __lsx_vpackev_b(alpha, out0); - dst0 = __lsx_vilvl_h(temp1, temp0); - dst1 = __lsx_vilvh_h(temp1, temp0); + dst0 = __lsx_vilvl_h(temp1, temp0); + dst1 = __lsx_vilvh_h(temp1, temp0); vec_yh = __lsx_vilvh_b(vec_y, vec_y); - y_ev = __lsx_vmulwev_w_hu_h(vec_yh, vec_yg); - y_od = __lsx_vmulwod_w_hu_h(vec_yh, vec_yg); - y_ev = __lsx_vsrai_w(y_ev, 16); - y_od = __lsx_vsrai_w(y_od, 16); - y_ev = __lsx_vadd_w(y_ev, vec_yb); - y_od = __lsx_vadd_w(y_od, vec_yb); - y_ev = __lsx_vsrai_w(y_ev, 6); - y_od = __lsx_vsrai_w(y_od, 6); - y_ev = __lsx_vclip255_w(y_ev); - y_od = __lsx_vclip255_w(y_od); - out0 = __lsx_vpackev_h(y_od, y_ev); + y_ev = __lsx_vmulwev_w_hu_h(vec_yh, vec_yg); + y_od = __lsx_vmulwod_w_hu_h(vec_yh, vec_yg); + y_ev = __lsx_vsrai_w(y_ev, 16); + y_od = __lsx_vsrai_w(y_od, 16); + y_ev = __lsx_vadd_w(y_ev, vec_yb); + y_od = __lsx_vadd_w(y_od, vec_yb); + y_ev = __lsx_vsrai_w(y_ev, 6); + y_od = __lsx_vsrai_w(y_od, 6); + y_ev = __lsx_vclip255_w(y_ev); + y_od = __lsx_vclip255_w(y_od); + out0 = __lsx_vpackev_h(y_od, y_ev); temp0 = __lsx_vpackev_b(out0, out0); temp1 = __lsx_vpackev_b(alpha, out0); - dst2 = __lsx_vilvl_h(temp1, temp0); - dst3 = __lsx_vilvh_h(temp1, temp0); + dst2 = __lsx_vilvl_h(temp1, temp0); + dst3 = __lsx_vilvh_h(temp1, temp0); __lsx_vst(dst0, dst_argb, 0); __lsx_vst(dst1, dst_argb, 16); __lsx_vst(dst2, dst_argb, 32); @@ -1360,7 +1371,7 @@ void YUY2ToARGBRow_LSX(const uint8_t* src_yuy2, __m128i vec_vrub, vec_vgug; __m128i out_b, out_g, out_r; __m128i const_80 = __lsx_vldi(0x480); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); __m128i alpha = __lsx_vldi(0xFF); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); @@ -1369,10 +1380,10 @@ void YUY2ToARGBRow_LSX(const uint8_t* src_yuy2, for (x = 0; x < len; x++) { src0 = __lsx_vld(src_yuy2, 0); - vec_y = __lsx_vpickev_b(src0, src0); + vec_y = __lsx_vpickev_b(src0, src0); vec_vu = __lsx_vpickod_b(src0, src0); - YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, - out_b, out_g, out_r); + YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, out_b, out_g, + out_r); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); src_yuy2 += 16; } @@ -1389,7 +1400,7 @@ void UYVYToARGBRow_LSX(const uint8_t* src_uyvy, __m128i vec_vrub, vec_vgug; __m128i out_b, out_g, out_r; __m128i const_80 = __lsx_vldi(0x480); - __m128i zero = __lsx_vldi(0); + __m128i zero = __lsx_vldi(0); __m128i alpha = __lsx_vldi(0xFF); YUVTORGB_SETUP(yuvconstants, vec_vr, vec_ub, vec_vg, vec_ug, vec_yg, vec_yb); @@ -1398,10 +1409,10 @@ void UYVYToARGBRow_LSX(const uint8_t* src_uyvy, for (x = 0; x < len; x++) { src0 = __lsx_vld(src_uyvy, 0); - vec_y = __lsx_vpickod_b(src0, src0); + vec_y = __lsx_vpickod_b(src0, src0); vec_vu = __lsx_vpickev_b(src0, src0); - YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, - out_b, out_g, out_r); + YUVTORGB(vec_y, vec_vu, vec_vrub, vec_vgug, vec_yg, vec_yb, out_b, out_g, + out_r); STOREARGB(alpha, out_r, out_g, out_b, dst_argb); src_uyvy += 16; } @@ -1535,8 +1546,8 @@ void ARGBExtractAlphaRow_LSX(const uint8_t* src_argb, __m128i src0, src1, src2, src3, tmp0, tmp1, dst0; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, - src_argb, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, src_argb, 48, + src0, src1, src2, src3); tmp0 = __lsx_vpickod_b(src1, src0); tmp1 = __lsx_vpickod_b(src3, src2); dst0 = __lsx_vpickod_b(tmp1, tmp0); @@ -1562,22 +1573,22 @@ void ARGBBlendRow_LSX(const uint8_t* src_argb, __m128i control = {0xFF000000FF000000, 0xFF000000FF000000}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, - src_argb1, 0, src_argb1, 16, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb1, 0, src_argb1, 16, + src0, src1, src2, src3); tmp0 = __lsx_vshuf4i_b(src0, 0xFF); tmp1 = __lsx_vshuf4i_b(src1, 0xFF); - a0 = __lsx_vilvl_b(zero, tmp0); - a1 = __lsx_vilvh_b(zero, tmp0); - a2 = __lsx_vilvl_b(zero, tmp1); - a3 = __lsx_vilvh_b(zero, tmp1); + a0 = __lsx_vilvl_b(zero, tmp0); + a1 = __lsx_vilvh_b(zero, tmp0); + a2 = __lsx_vilvl_b(zero, tmp1); + a3 = __lsx_vilvh_b(zero, tmp1); reg0 = __lsx_vilvl_b(zero, src2); reg1 = __lsx_vilvh_b(zero, src2); reg2 = __lsx_vilvl_b(zero, src3); reg3 = __lsx_vilvh_b(zero, src3); DUP4_ARG2(__lsx_vsub_h, const_256, a0, const_256, a1, const_256, a2, const_256, a3, a0, a1, a2, a3); - DUP4_ARG2(__lsx_vmul_h, a0, reg0, a1, reg1, a2, reg2, a3, reg3, - reg0, reg1, reg2, reg3); + DUP4_ARG2(__lsx_vmul_h, a0, reg0, a1, reg1, a2, reg2, a3, reg3, reg0, reg1, + reg2, reg3); DUP2_ARG3(__lsx_vsrani_b_h, reg1, reg0, 8, reg3, reg2, 8, dst0, dst1); dst0 = __lsx_vsadd_bu(dst0, src0); dst1 = __lsx_vsadd_bu(dst1, src1); @@ -1608,8 +1619,8 @@ void ARGBQuantizeRow_LSX(uint8_t* dst_argb, __m128i control = {0xFF000000FF000000, 0xFF000000FF000000}; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, dst_argb, 0, dst_argb, 16, dst_argb, 32, - dst_argb, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, dst_argb, 0, dst_argb, 16, dst_argb, 32, dst_argb, 48, + src0, src1, src2, src3); reg0 = __lsx_vilvl_b(zero, src0); reg1 = __lsx_vilvh_b(zero, src0); reg2 = __lsx_vilvl_b(zero, src1); @@ -1652,10 +1663,10 @@ void ARGBQuantizeRow_LSX(uint8_t* dst_argb, dst3 = __lsx_vpickev_b(reg3, reg2); DUP4_ARG2(__lsx_vmul_b, dst0, vec_size, dst1, vec_size, dst2, vec_size, dst3, vec_size, dst0, dst1, dst2, dst3); - DUP4_ARG2(__lsx_vadd_b, dst0, vec_offset, dst1, vec_offset, dst2, vec_offset, - dst3, vec_offset, dst0, dst1, dst2, dst3); - DUP4_ARG3(__lsx_vbitsel_v, dst0, src0, control, dst1, src1, control, - dst2, src2, control, dst3, src3, control, dst0, dst1, dst2, dst3); + DUP4_ARG2(__lsx_vadd_b, dst0, vec_offset, dst1, vec_offset, dst2, + vec_offset, dst3, vec_offset, dst0, dst1, dst2, dst3); + DUP4_ARG3(__lsx_vbitsel_v, dst0, src0, control, dst1, src1, control, dst2, + src2, control, dst3, src3, control, dst0, dst1, dst2, dst3); __lsx_vst(dst0, dst_argb, 0); __lsx_vst(dst1, dst_argb, 16); __lsx_vst(dst2, dst_argb, 32); @@ -1684,22 +1695,24 @@ void ARGBColorMatrixRow_LSX(const uint8_t* src_argb, src0, matrix_a, tmp_b, tmp_g, tmp_r, tmp_a); DUP4_ARG2(__lsx_vdp2_h_bu_b, src1, matrix_b, src1, matrix_g, src1, matrix_r, src1, matrix_a, reg_b, reg_g, reg_r, reg_a); - DUP4_ARG2(__lsx_vhaddw_w_h, tmp_b, tmp_b, tmp_g, tmp_g, tmp_r, tmp_r, - tmp_a, tmp_a, tmp_b, tmp_g, tmp_r, tmp_a); - DUP4_ARG2(__lsx_vhaddw_w_h, reg_b, reg_b, reg_g, reg_g, reg_r, reg_r, - reg_a, reg_a, reg_b, reg_g, reg_r, reg_a); - DUP4_ARG2(__lsx_vsrai_w, tmp_b, 6, tmp_g, 6, tmp_r, 6, - tmp_a, 6, tmp_b, tmp_g, tmp_r, tmp_a); - DUP4_ARG2(__lsx_vsrai_w, reg_b, 6, reg_g, 6, reg_r, 6, - reg_a, 6, reg_b, reg_g, reg_r, reg_a); - DUP4_ARG1(__lsx_vclip255_w, tmp_b, tmp_g, tmp_r, tmp_a, tmp_b, tmp_g, tmp_r, tmp_a) - DUP4_ARG1(__lsx_vclip255_w, reg_b, reg_g, reg_r, reg_a, reg_b, reg_g, reg_r, reg_a) - DUP4_ARG2(__lsx_vpickev_h, reg_b, tmp_b, reg_g, tmp_g, reg_r, tmp_r, - reg_a, tmp_a, tmp_b, tmp_g, tmp_r, tmp_a); - tmp0 = __lsx_vpackev_b(tmp_g, tmp_b); - tmp1 = __lsx_vpackev_b(tmp_a, tmp_r); - dst0 = __lsx_vilvl_h(tmp1, tmp0); - dst1 = __lsx_vilvh_h(tmp1, tmp0); + DUP4_ARG2(__lsx_vhaddw_w_h, tmp_b, tmp_b, tmp_g, tmp_g, tmp_r, tmp_r, tmp_a, + tmp_a, tmp_b, tmp_g, tmp_r, tmp_a); + DUP4_ARG2(__lsx_vhaddw_w_h, reg_b, reg_b, reg_g, reg_g, reg_r, reg_r, reg_a, + reg_a, reg_b, reg_g, reg_r, reg_a); + DUP4_ARG2(__lsx_vsrai_w, tmp_b, 6, tmp_g, 6, tmp_r, 6, tmp_a, 6, tmp_b, + tmp_g, tmp_r, tmp_a); + DUP4_ARG2(__lsx_vsrai_w, reg_b, 6, reg_g, 6, reg_r, 6, reg_a, 6, reg_b, + reg_g, reg_r, reg_a); + DUP4_ARG1(__lsx_vclip255_w, tmp_b, tmp_g, tmp_r, tmp_a, tmp_b, tmp_g, tmp_r, + tmp_a) + DUP4_ARG1(__lsx_vclip255_w, reg_b, reg_g, reg_r, reg_a, reg_b, reg_g, reg_r, + reg_a) + DUP4_ARG2(__lsx_vpickev_h, reg_b, tmp_b, reg_g, tmp_g, reg_r, tmp_r, reg_a, + tmp_a, tmp_b, tmp_g, tmp_r, tmp_a); + tmp0 = __lsx_vpackev_b(tmp_g, tmp_b); + tmp1 = __lsx_vpackev_b(tmp_a, tmp_r); + dst0 = __lsx_vilvl_h(tmp1, tmp0); + dst1 = __lsx_vilvh_h(tmp1, tmp0); __lsx_vst(dst0, dst_argb, 0); __lsx_vst(dst1, dst_argb, 16); src_argb += 32; @@ -1717,8 +1730,8 @@ void SplitUVRow_LSX(const uint8_t* src_uv, __m128i dst0, dst1, dst2, dst3; for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src_uv, 0, src_uv, 16, src_uv, 32, - src_uv, 48, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vld, src_uv, 0, src_uv, 16, src_uv, 32, src_uv, 48, src0, + src1, src2, src3); DUP2_ARG2(__lsx_vpickev_b, src1, src0, src3, src2, dst0, dst1); DUP2_ARG2(__lsx_vpickod_b, src1, src0, src3, src2, dst2, dst3); __lsx_vst(dst0, dst_u, 0); @@ -1756,10 +1769,10 @@ void MirrorSplitUVRow_LSX(const uint8_t* src_uv, src_uv += (width << 1); for (x = 0; x < len; x++) { src_uv -= 64; - DUP4_ARG2(__lsx_vld, src_uv, 0, src_uv, 16, src_uv, 32, - src_uv, 48, src2, src3, src0, src1); - DUP4_ARG3(__lsx_vshuf_b, src1, src0, shuff1, src3, src2, shuff1, - src1, src0, shuff0, src3, src2, shuff0, dst0, dst1, dst2, dst3); + DUP4_ARG2(__lsx_vld, src_uv, 0, src_uv, 16, src_uv, 32, src_uv, 48, src2, + src3, src0, src1); + DUP4_ARG3(__lsx_vshuf_b, src1, src0, shuff1, src3, src2, shuff1, src1, src0, + shuff0, src3, src2, shuff0, dst0, dst1, dst2, dst3); __lsx_vst(dst0, dst_v, 0); __lsx_vst(dst1, dst_v, 16); __lsx_vst(dst2, dst_u, 0); @@ -1778,18 +1791,21 @@ void HalfFloatRow_LSX(const uint16_t* src, float mult = 1.9259299444e-34f * scale; __m128i src0, src1, src2, src3, dst0, dst1, dst2, dst3; __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; - __m128 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7; + __m128 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7; __m128 vec_mult = (__m128)__lsx_vldrepl_w(&mult, 0); __m128i zero = __lsx_vldi(0); for (x = 0; x < len; x++) { - DUP4_ARG2(__lsx_vld, src, 0, src, 16, src, 32, src, 48, src0, src1, src2, src3); - DUP4_ARG2(__lsx_vilvl_h, zero, src0, zero, src1, zero, src2, - zero, src3, tmp0, tmp2, tmp4, tmp6); - DUP4_ARG2(__lsx_vilvh_h, zero, src0, zero, src1, zero, src2, - zero, src3, tmp1, tmp3, tmp5, tmp7); - DUP4_ARG1(__lsx_vffint_s_wu, tmp0, tmp2, tmp4, tmp6, reg0, reg2, reg4, reg6); - DUP4_ARG1(__lsx_vffint_s_wu, tmp1, tmp3, tmp5, tmp7, reg1, reg3, reg5, reg7); + DUP4_ARG2(__lsx_vld, src, 0, src, 16, src, 32, src, 48, src0, src1, src2, + src3); + DUP4_ARG2(__lsx_vilvl_h, zero, src0, zero, src1, zero, src2, zero, src3, + tmp0, tmp2, tmp4, tmp6); + DUP4_ARG2(__lsx_vilvh_h, zero, src0, zero, src1, zero, src2, zero, src3, + tmp1, tmp3, tmp5, tmp7); + DUP4_ARG1(__lsx_vffint_s_wu, tmp0, tmp2, tmp4, tmp6, reg0, reg2, reg4, + reg6); + DUP4_ARG1(__lsx_vffint_s_wu, tmp1, tmp3, tmp5, tmp7, reg1, reg3, reg5, + reg7); DUP4_ARG2(__lsx_vfmul_s, reg0, vec_mult, reg1, vec_mult, reg2, vec_mult, reg3, vec_mult, reg0, reg1, reg2, reg3); DUP4_ARG2(__lsx_vfmul_s, reg4, vec_mult, reg5, vec_mult, reg6, vec_mult, @@ -1798,8 +1814,8 @@ void HalfFloatRow_LSX(const uint16_t* src, (v4u32)reg3, 13, tmp0, tmp1, tmp2, tmp3); DUP4_ARG2(__lsx_vsrli_w, (v4u32)reg4, 13, (v4u32)reg5, 13, (v4u32)reg6, 13, (v4u32)reg7, 13, tmp4, tmp5, tmp6, tmp7); - DUP4_ARG2(__lsx_vpickev_h, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, - tmp7, tmp6, dst0, dst1, dst2, dst3); + DUP4_ARG2(__lsx_vpickev_h, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, tmp7, tmp6, + dst0, dst1, dst2, dst3); __lsx_vst(dst0, dst, 0); __lsx_vst(dst1, dst, 16); __lsx_vst(dst2, dst, 32); |