[FFmpeg-devel] [PATCHv3 1/3] sws/input: R-V V rgb24ToY & bgr24ToY

[FFmpeg-devel] [PATCHv3 1/3] sws/input: R-V V rgb24ToY & bgr24ToY

[Rémi Denis-Courmont]

T-Head C908:
rgb24_to_y_8_c:            2.0
rgb24_to_y_8_rvv_i32:      2.7
rgb24_to_y_128_c:         26.2
rgb24_to_y_128_rvv_i32:    9.2
rgb24_to_y_1080_c:       219.5
rgb24_to_y_1080_rvv_i32:  76.2
rgb24_to_y_1280_c:       276.2
rgb24_to_y_1280_rvv_i32:  89.7
rgb24_to_y_1920_c:       389.7
rgb24_to_y_1920_rvv_i32: 134.2

SpacemiT X60:
rgb24_to_y_8_c:            1.7
rgb24_to_y_8_rvv_i32:      2.2
rgb24_to_y_128_c:         23.2
rgb24_to_y_128_rvv_i32:    4.2
rgb24_to_y_1080_c:       195.0
rgb24_to_y_1080_rvv_i32:  33.7
rgb24_to_y_1280_c:       231.0
rgb24_to_y_1280_rvv_i32:  40.0
rgb24_to_y_1920_c:       346.2
rgb24_to_y_1920_rvv_i32:  59.7
---
 libswscale/riscv/Makefile     |  6 ++--
 libswscale/riscv/input.S      | 55 +++++++++++++++++++++++++++++++++++
 libswscale/riscv/swscale.c    | 46 +++++++++++++++++++++++++++++
 libswscale/swscale.c          |  2 ++
 libswscale/swscale_internal.h |  1 +
 5 files changed, 108 insertions(+), 2 deletions(-)
 create mode 100644 libswscale/riscv/input.S
 create mode 100644 libswscale/riscv/swscale.c

diff --git a/libswscale/riscv/Makefile b/libswscale/riscv/Makefile
index 48afaf62aa..d1b0eae367 100644
--- a/libswscale/riscv/Makefile
+++ b/libswscale/riscv/Makefile
@@ -1,3 +1,5 @@
-OBJS += riscv/rgb2rgb.o
+OBJS += riscv/rgb2rgb.o \
+        riscv/swscale.o
 RV-OBJS += riscv/rgb2rgb_rvb.o
-RVV-OBJS += riscv/rgb2rgb_rvv.o
+RVV-OBJS += riscv/input.o \
+            riscv/rgb2rgb_rvv.o
diff --git a/libswscale/riscv/input.S b/libswscale/riscv/input.S
new file mode 100644
index 0000000000..323f650bc9
--- /dev/null
+++ b/libswscale/riscv/input.S
@@ -0,0 +1,55 @@
+/*
+ * Copyright © 2024 Rémi Denis-Courmont.
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/riscv/asm.S"
+
+func ff_bgr24ToY_rvv, zve32x
+        lw      t1, 8(a5) # BY
+        lw      t3, 0(a5) # RY
+        j       1f
+endfunc
+
+func ff_rgb24ToY_rvv, zve32x
+        lw      t1, 0(a5) # RY
+        lw      t3, 8(a5) # BY
+1:
+        lw      t2, 4(a5) # GY
+        li      t4, (32 << (15 - 1)) + (1 << (15 - 7))
+2:
+        vsetvli    t0, a4, e32, m8, ta, ma
+        vlseg3e8.v v0, (a1)
+        sub        a4, a4, t0
+        vzext.vf4  v8, v0
+        sh1add     t5, t0, t0 # t1 = 3 * t0
+        vzext.vf4  v16, v2
+        vzext.vf4  v24, v4
+        add        a1, t5, a1
+        vmul.vx    v8, v8, t1
+        vmacc.vx   v8, t2, v16
+        vmacc.vx   v8, t3, v24
+        vadd.vx    v8, v8, t4
+        vsetvli    zero, zero, e16, m4, ta, ma
+        vnsra.wi   v0, v8, 15 - 6
+        vse16.v    v0, (a0)
+        sh1add     a0, t0, a0
+        bnez       a4, 2b
+
+        ret
+endfunc
diff --git a/libswscale/riscv/swscale.c b/libswscale/riscv/swscale.c
new file mode 100644
index 0000000000..187b3fce58
--- /dev/null
+++ b/libswscale/riscv/swscale.c
@@ -0,0 +1,46 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/riscv/cpu.h"
+#include "libswscale/swscale_internal.h"
+
+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_rgb24ToY_rvv(uint8_t *dst, const uint8_t *src, const uint8_t *,
+                     const uint8_t *, int width, uint32_t *coeffs, void *);
+
+av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
+{
+#if HAVE_RVV
+    int flags = av_get_cpu_flags();
+
+    if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
+        switch (c->srcFormat) {
+            case AV_PIX_FMT_BGR24:
+                c->lumToYV12 = ff_bgr24ToY_rvv;
+                break;
+
+            case AV_PIX_FMT_RGB24:
+                c->lumToYV12 = ff_rgb24ToY_rvv;
+                break;
+        }
+    }
+#endif
+}
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 2795429b6c..9736734881 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -602,6 +602,8 @@ void ff_sws_init_scale(SwsContext *c)
     ff_sws_init_swscale_arm(c);
 #elif ARCH_LOONGARCH64
     ff_sws_init_swscale_loongarch(c);
+#elif ARCH_RISCV
+    ff_sws_init_swscale_riscv(c);
 #endif
 }
 
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index d4b0c3cee2..5007dd422f 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -988,6 +988,7 @@ void ff_sws_init_swscale_x86(SwsContext *c);
 void ff_sws_init_swscale_aarch64(SwsContext *c);
 void ff_sws_init_swscale_arm(SwsContext *c);
 void ff_sws_init_swscale_loongarch(SwsContext *c);
+void ff_sws_init_swscale_riscv(SwsContext *c);
 
 void ff_hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
                        const uint8_t *src, int srcW, int xInc);
-- 
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".

[FFmpeg-devel] [PATCH 2/3] sws/input: R-V V rgb24ToUV and bgr24ToUV

[Rémi Denis-Courmont]

T-Head C908:
rgb24_to_uv_8_c:            2.7
rgb24_to_uv_8_rvv_i32:      3.2
rgb24_to_uv_128_c:         41.0
rgb24_to_uv_128_rvv_i32:   12.7
rgb24_to_uv_1080_c:       342.5
rgb24_to_uv_1080_rvv_i32: 105.7
rgb24_to_uv_1280_c:       406.0
rgb24_to_uv_1280_rvv_i32: 124.2
rgb24_to_uv_1920_c:       626.0
rgb24_to_uv_1920_rvv_i32: 186.0

SpacemiT X60:
rgb24_to_uv_8_c:            2.5
rgb24_to_uv_8_rvv_i32:      3.0
rgb24_to_uv_128_c:         36.5
rgb24_to_uv_128_rvv_i32:    5.7
rgb24_to_uv_1080_c:       304.2
rgb24_to_uv_1080_rvv_i32:  49.0
rgb24_to_uv_1280_c:       360.5
rgb24_to_uv_1280_rvv_i32:  57.5
rgb24_to_uv_1920_c:       540.7
rgb24_to_uv_1920_rvv_i32:  86.2
---
 libswscale/riscv/input.S   | 46 ++++++++++++++++++++++++++++++++++++++
 libswscale/riscv/swscale.c |  8 +++++++
 2 files changed, 54 insertions(+)

diff --git a/libswscale/riscv/input.S b/libswscale/riscv/input.S
index 323f650bc9..3392f189ca 100644
--- a/libswscale/riscv/input.S
+++ b/libswscale/riscv/input.S
@@ -53,3 +53,49 @@ func ff_rgb24ToY_rvv, zve32x
 
         ret
 endfunc
+
+func ff_bgr24ToUV_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_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)) + (1 << (15 - 7))
+2:
+        vsetvli    t0, a5, e32, m8, ta, ma
+        vlseg3e8.v v0, (a3)
+        sub        a5, a5, t0
+        vzext.vf4  v16, v0
+        sh1add     a6, t0, t0
+        vzext.vf4  v24, v2
+        vmul.vx    v8, v16, t1
+        add        a3, a6, a3
+        vmul.vx    v16, v16, t4
+        vmacc.vx   v8, t2, v24
+        vmacc.vx   v16, t5, v24
+        vzext.vf4  v24, v4
+        vadd.vx    v8, v8, a7
+        vadd.vx    v16, v16, a7
+        vmacc.vx   v8, t3, v24
+        vmacc.vx   v16, t6, v24
+        vsetvli    zero, zero, e16, m4, ta, ma
+        vnsra.wi   v0, v8, 15 - 6
+        vnsra.wi   v4, v16, 15 - 6
+        vse16.v    v0, (a0)
+        sh1add     a0, t0, a0
+        vse16.v    v4, (a1)
+        sh1add     a1, t0, a1
+        bnez       a5, 2b
+
+        ret
+endfunc
diff --git a/libswscale/riscv/swscale.c b/libswscale/riscv/swscale.c
index 187b3fce58..b3552976c6 100644
--- a/libswscale/riscv/swscale.c
+++ b/libswscale/riscv/swscale.c
@@ -23,8 +23,12 @@
 
 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_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 *);
 
 av_cold void ff_sws_init_swscale_riscv(SwsContext *c)
 {
@@ -35,10 +39,14 @@ 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)
+                    c->chrToYV12 = ff_bgr24ToUV_rvv;
                 break;
 
             case AV_PIX_FMT_RGB24:
                 c->lumToYV12 = ff_rgb24ToY_rvv;
+                if (!c->chrSrcHSubSample)
+                    c->chrToYV12 = ff_rgb24ToUV_rvv;
                 break;
         }
     }
-- 
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".

[FFmpeg-devel] [PATCH 3/3] sws/input: R-V V rgb24ToUV_half and bgr24ToUV_half

[Rémi Denis-Courmont]

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

diff --git a/libswscale/riscv/input.S b/libswscale/riscv/input.S
index 3392f189ca..f84f79899b 100644
--- a/libswscale/riscv/input.S
+++ b/libswscale/riscv/input.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;
         }
-- 
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".