[FFmpeg-devel] [PATCH 1/1] lavc/huffyuvdsp: basic R-V V add_hfyu_left_pred_bgr32

[Rémi Denis-Courmont]

Better performance can probably be achieved with a more intricate
unrolled loop, but this is a start:

add_hfyu_left_pred_bgr32_c: 15084.0
add_hfyu_left_pred_bgr32_rvv_i32: 10280.2

This would actually be cleaner with the RISC-V P extension, but that is
not ratified yet (I think?) and usually not supported if V is supported.
---
 libavcodec/riscv/huffyuvdsp_init.c |  6 +++++-
 libavcodec/riscv/huffyuvdsp_rvv.S  | 16 ++++++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/libavcodec/riscv/huffyuvdsp_init.c b/libavcodec/riscv/huffyuvdsp_init.c
index 115b25881c..b49b3dc097 100644
--- a/libavcodec/riscv/huffyuvdsp_init.c
+++ b/libavcodec/riscv/huffyuvdsp_init.c
@@ -24,6 +24,8 @@
 #include "libavcodec/huffyuvdsp.h"
 
 void ff_add_int16_rvv(uint16_t *dst, const uint16_t *src, unsigned m, int w);
+void ff_add_hfyu_left_pred_bgr32_rvv(uint8_t *dst, const uint8_t *src,
+                                     intptr_t w, uint8_t *left);
 
 av_cold void ff_huffyuvdsp_init_riscv(HuffYUVDSPContext *c,
                                       enum AVPixelFormat pix_fmt)
@@ -31,7 +33,9 @@ av_cold void ff_huffyuvdsp_init_riscv(HuffYUVDSPContext *c,
 #if HAVE_RVV
     int flags = av_get_cpu_flags();
 
-    if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR))
+    if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
         c->add_int16 = ff_add_int16_rvv;
+        c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_rvv;
+    }
 #endif
 }
diff --git a/libavcodec/riscv/huffyuvdsp_rvv.S b/libavcodec/riscv/huffyuvdsp_rvv.S
index f8926fdaea..9c4434907d 100644
--- a/libavcodec/riscv/huffyuvdsp_rvv.S
+++ b/libavcodec/riscv/huffyuvdsp_rvv.S
@@ -35,3 +35,19 @@ func ff_add_int16_rvv, zve32x
 
         ret
 endfunc
+
+func ff_add_hfyu_left_pred_bgr32_rvv, zve32x
+        vsetivli zero, 4, e8, m1, ta, ma
+        vle8.v  v8, (a3)
+        sh2add  a2, a2, a1
+1:
+        vle8.v  v0, (a1)
+        vadd.vv v8, v8, v0
+        addi    a1, a1, 4
+        vse8.v  v8, (a0)
+        addi    a0, a0, 4
+        bne     a2, a1, 1b
+
+        vse8.v  v8, (a3)
+        ret
+endfunc
-- 
2.42.0