[FFmpeg-devel] [PATCH 3/3] sws/input: R-V V 32-bit RGB to halved UV

[Rémi Denis-Courmont]

T-Head C908:
abgr_to_uv_half_8_c:            2.2
abgr_to_uv_half_8_rvv_i32:      3.5
abgr_to_uv_half_128_c:         44.0
abgr_to_uv_half_128_rvv_i32:   13.0
abgr_to_uv_half_1080_c:       245.0
abgr_to_uv_half_1080_rvv_i32: 107.2
abgr_to_uv_half_1920_c:       406.2
abgr_to_uv_half_1920_rvv_i32: 188.7
bgra_to_uv_half_8_c:            2.2
bgra_to_uv_half_8_rvv_i32:      3.5
bgra_to_uv_half_128_c:         26.5
bgra_to_uv_half_128_rvv_i32:   13.0
bgra_to_uv_half_1080_c:       219.7
bgra_to_uv_half_1080_rvv_i32: 107.0
bgra_to_uv_half_1920_c:       406.7
bgra_to_uv_half_1920_rvv_i32: 188.7

SpacemiT X60:
abgr_to_uv_half_8_c:           2.2
abgr_to_uv_half_8_rvv_i32:     3.0
abgr_to_uv_half_128_c:        28.2
abgr_to_uv_half_128_rvv_i32:   5.7
abgr_to_uv_half_1080_c:      235.5
abgr_to_uv_half_1080_rvv_i32: 47.7
abgr_to_uv_half_1920_c:      418.2
abgr_to_uv_half_1920_rvv_i32: 84.0
bgra_to_uv_half_8_c:           2.0
bgra_to_uv_half_8_rvv_i32:     3.0
bgra_to_uv_half_128_c:        23.7
bgra_to_uv_half_128_rvv_i32:   5.7
bgra_to_uv_half_1080_c:      195.5
bgra_to_uv_half_1080_rvv_i32: 47.7
bgra_to_uv_half_1920_c:      346.5
bgra_to_uv_half_1920_rvv_i32: 84.0
---
 libswscale/riscv/input.S   | 61 ++++++++++++++++++++++++++++++++++++++
 libswscale/riscv/swscale.c | 16 +++++++---
 2 files changed, 73 insertions(+), 4 deletions(-)

diff --git a/libswscale/riscv/input.S b/libswscale/riscv/input.S
index 2d00a5a225..1d7de59c66 100644
--- a/libswscale/riscv/input.S
+++ b/libswscale/riscv/input.S
@@ -242,6 +242,67 @@ func ff_\chr0\()ToUV_rvv, zve32x
 
         ret
 endfunc
+
+func ff_\chr1\()ToUV_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_\chr0\()ToUV_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      a6, 0xff
+        li      a7, (256 << 15) + (1 << (15 - 6))
+2:
+        vsetvli     t0, a5, e32, m4, ta, ma
+        vlseg2e32.v v0, (a3)
+        sub         a5, a5, t0
+        .if     \high
+        vsrl.vi     v8, v0, 24
+        vsrl.vi     v12, v4, 24
+        .else
+        vand.vx     v8, v0, a6
+        vand.vx     v12, v4, a6
+        .endif
+        sh3add      a3, t0, a3
+        vsrl.vi     v16, v0, 8 * (1 + \high)
+        vsrl.vi     v20, v4, 8 * (1 + \high)
+        vsrl.vi     v24, v0, 8 * (2 - \high)
+        vsrl.vi     v28, v4, 8 * (2 - \high)
+        vand.vx     v16, v16, a6
+        vand.vx     v20, v20, a6
+        vand.vx     v24, v24, a6
+        vand.vx     v28, v28, a6
+        vadd.vv     v8, v8, v12
+        vadd.vv     v16, v16, v20
+        vadd.vv     v24, v24, v28
+        vmul.vx     v0, v8, t1
+        vmul.vx     v4, v8, t4
+        vmacc.vx    v0, t2, v16
+        vmacc.vx    v4, t5, v16
+        vmacc.vx    v0, t3, v24
+        vmacc.vx    v4, t6, v24
+        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
         .endm
 
 rgba_input rgba32, bgra32, 0
diff --git a/libswscale/riscv/swscale.c b/libswscale/riscv/swscale.c
index 76fbf2ca08..ffeb7876bf 100644
--- a/libswscale/riscv/swscale.c
+++ b/libswscale/riscv/swscale.c
@@ -47,13 +47,17 @@ av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
         switch (c->srcFormat) {
             case AV_PIX_FMT_ABGR:
                 c->lumToYV12 = ff_abgr32ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_abgr32ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_abgr32ToUV_rvv;
                 break;
 
             case AV_PIX_FMT_ARGB:
                 c->lumToYV12 = ff_argb32ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_argb32ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_argb32ToUV_rvv;
                 break;
 
@@ -67,7 +71,9 @@ av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
 
             case AV_PIX_FMT_BGRA:
                 c->lumToYV12 = ff_bgra32ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_bgra32ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_bgra32ToUV_rvv;
                 break;
 
@@ -81,7 +87,9 @@ av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
 
             case AV_PIX_FMT_RGBA:
                 c->lumToYV12 = ff_rgba32ToY_rvv;
-                if (!c->chrSrcHSubSample)
+                if (c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_rgba32ToUV_half_rvv;
+                else
                     c->chrToYV12 = ff_rgba32ToUV_rvv;
                 break;
         }
-- 
2.45.1