[FFmpeg-cvslog] sws/input: R-V V rgb24ToUV_half and bgr24ToUV_half

[Rémi Denis-Courmont]

ffmpeg | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Wed Jun  5 19:32:31 2024 +0300| [e0f4d185f1ea12e9a7bfa078a4224687102751e7] | committer: Rémi Denis-Courmont

sws/input: R-V V rgb24ToUV_half and bgr24ToUV_half

T-Head C908:
rgb24_to_uv_half_4_c:           2.0
rgb24_to_uv_half_4_rvv_i32:     3.5
rgb24_to_uv_half_64_c:         27.0
rgb24_to_uv_half_64_rvv_i32:   12.5
rgb24_to_uv_half_540_c:       223.7
rgb24_to_uv_half_540_rvv_i32: 105.2
rgb24_to_uv_half_640_c:       265.5
rgb24_to_uv_half_640_rvv_i32: 123.7
rgb24_to_uv_half_960_c:       414.5
rgb24_to_uv_half_960_rvv_i32: 249.5

SpacemiT X60:
rgb24_to_uv_half_4_c:           1.7
rgb24_to_uv_half_4_rvv_i32:     4.2
rgb24_to_uv_half_64_c:         24.0
rgb24_to_uv_half_64_rvv_i32:    8.7
rgb24_to_uv_half_540_c:       199.2
rgb24_to_uv_half_540_rvv_i32:  72.5
rgb24_to_uv_half_640_c:       235.7
rgb24_to_uv_half_640_rvv_i32:  85.2
rgb24_to_uv_half_960_c:       353.5
rgb24_to_uv_half_960_rvv_i32: 127.5

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=e0f4d185f1ea12e9a7bfa078a4224687102751e7
---

 libswscale/riscv/input_rvv.S | 50 ++++++++++++++++++++++++++++++++++++++++++++
 libswscale/riscv/swscale.c   | 14 +++++++++++--
 2 files changed, 62 insertions(+), 2 deletions(-)

diff --git a/libswscale/riscv/input_rvv.S b/libswscale/riscv/input_rvv.S
index 3392f189ca..f84f79899b 100644
--- a/libswscale/riscv/input_rvv.S
+++ b/libswscale/riscv/input_rvv.S
@@ -99,3 +99,53 @@ func ff_rgb24ToUV_rvv, zve32x
 
         ret
 endfunc
+
+func ff_bgr24ToUV_half_rvv, zve32x
+        lw      t1, 20(a6) # BU
+        lw      t4, 32(a6) # BV
+        lw      t3, 12(a6) # RU
+        lw      t6, 24(a6) # RV
+        j       1f
+endfunc
+
+func ff_rgb24ToUV_half_rvv, zve32x
+        lw      t1, 12(a6) # RU
+        lw      t4, 24(a6) # RV
+        lw      t3, 20(a6) # BU
+        lw      t6, 32(a6) # BV
+1:
+        lw      t2, 16(a6) # GU
+        lw      t5, 28(a6) # GV
+        li      a7, (256 << 15) + (1 << (15 - 6))
+2:
+        vsetvli    t0, a5, e8, m1, ta, ma
+        vlseg6e8.v v0, (a3)
+        sh1add     a6, t0, t0
+        vwaddu.vv  v8, v0, v3
+        sub        a5, a5, t0
+        vwaddu.vv  v10, v1, v4
+        sh1add     a3, a6, a3
+        vwaddu.vv  v12, v2, v5
+        vsetvli    zero, zero, e32, m4, ta, ma
+        vzext.vf2  v20, v8
+        vzext.vf2  v24, v10
+        vzext.vf2  v28, v12
+        vmul.vx    v0, v20, t1
+        vmul.vx    v4, v20, t4
+        vmacc.vx   v0, t2, v24
+        vmacc.vx   v4, t5, v24
+        vmacc.vx   v0, t3, v28
+        vmacc.vx   v4, t6, v28
+        vadd.vx    v0, v0, a7
+        vadd.vx    v4, v4, a7
+        vsetvli    zero, zero, e16, m2, ta, ma
+        vnsra.wi   v0, v0, 15 - 5
+        vnsra.wi   v2, v4, 15 - 5
+        vse16.v    v0, (a0)
+        sh1add     a0, t0, a0
+        vse16.v    v2, (a1)
+        sh1add     a1, t0, a1
+        bnez       a5, 2b
+
+        ret
+endfunc
diff --git a/libswscale/riscv/swscale.c b/libswscale/riscv/swscale.c
index b3552976c6..bae626f64a 100644
--- a/libswscale/riscv/swscale.c
+++ b/libswscale/riscv/swscale.c
@@ -25,10 +25,16 @@ void ff_bgr24ToY_rvv(uint8_t *dst, const uint8_t *src, const uint8_t *,
                      const uint8_t *, int width, uint32_t *coeffs, void *);
 void ff_bgr24ToUV_rvv(uint8_t *, uint8_t *, const uint8_t *, const uint8_t *,
                       const uint8_t *, int width, uint32_t *coeffs, void *);
+void ff_bgr24ToUV_half_rvv(uint8_t *, uint8_t *, const uint8_t *,
+                           const uint8_t *, const uint8_t *, int width,
+                           uint32_t *coeffs, void *);
 void ff_rgb24ToY_rvv(uint8_t *dst, const uint8_t *src, const uint8_t *,
                      const uint8_t *, int width, uint32_t *coeffs, void *);
 void ff_rgb24ToUV_rvv(uint8_t *, uint8_t *, const uint8_t *, const uint8_t *,
                       const uint8_t *, int width, uint32_t *coeffs, void *);
+void ff_rgb24ToUV_half_rvv(uint8_t *, uint8_t *, const uint8_t *,
+                           const uint8_t *, const uint8_t *, int width,
+                           uint32_t *coeffs, void *);
 
 av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
 {
@@ -39,13 +45,17 @@ av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
         switch (c->srcFormat) {
             case AV_PIX_FMT_BGR24:
                 c->lumToYV12 = ff_bgr24ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_bgr24ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_bgr24ToUV_rvv;
                 break;
 
             case AV_PIX_FMT_RGB24:
                 c->lumToYV12 = ff_rgb24ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_rgb24ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_rgb24ToUV_rvv;
                 break;
         }