[FFmpeg-devel] [PATCH 1/4] lavu/fixed_dsp: R-V V vector_fmul_add

[Rémi Denis-Courmont]

vector_fmul_add_fixed_c: 2.2
vector_fmul_add_fixed_rvv_i64: 0.5
---
 libavutil/riscv/fixed_dsp_init.c |  5 +++++
 libavutil/riscv/fixed_dsp_rvv.S  | 21 +++++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/libavutil/riscv/fixed_dsp_init.c b/libavutil/riscv/fixed_dsp_init.c
index 409b6d7b55..4b239829f5 100644
--- a/libavutil/riscv/fixed_dsp_init.c
+++ b/libavutil/riscv/fixed_dsp_init.c
@@ -25,6 +25,8 @@
 #include "libavutil/cpu.h"
 #include "libavutil/fixed_dsp.h"
 
+void ff_vector_fmul_add_fixed_rvv(int *dst, const int *src0, const int *src1,
+                                  const int *src2, int len);
 int ff_scalarproduct_fixed_rvv(const int *v1, const int *v2, int len);
 void ff_butterflies_fixed_rvv(int *v1, int *v2, int len);
 
@@ -34,8 +36,11 @@ av_cold void ff_fixed_dsp_init_riscv(AVFixedDSPContext *fdsp)
     int flags = av_get_cpu_flags();
 
     if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
+        fdsp->vector_fmul_add = ff_vector_fmul_add_fixed_rvv;
+
         if (flags & AV_CPU_FLAG_RVV_I64)
             fdsp->scalarproduct_fixed = ff_scalarproduct_fixed_rvv;
+
         fdsp->butterflies_fixed = ff_butterflies_fixed_rvv;
     }
 #endif
diff --git a/libavutil/riscv/fixed_dsp_rvv.S b/libavutil/riscv/fixed_dsp_rvv.S
index 83cad4bde3..1022a401d3 100644
--- a/libavutil/riscv/fixed_dsp_rvv.S
+++ b/libavutil/riscv/fixed_dsp_rvv.S
@@ -20,6 +20,27 @@
 
 #include "asm.S"
 
+
+func ff_vector_fmul_add_fixed_rvv, zve32x
+        csrwi   vxrm, 0
+1:
+        vsetvli t0, a4, e32, m8, ta, ma
+        vle32.v v16, (a1)
+        sub     a4, a4, t0
+        vle32.v v24, (a2)
+        sh2add  a1, t0, a1
+        vsmul.vv v8, v16, v24
+        sh2add  a2, t0, a2
+        vle32.v v0,(a3)
+        sh2add  a3, t0, a3
+        vadd.vv v8, v8, v0
+        vse32.v v8, (a0)
+        sh2add  a0, t0, a0
+        bnez    a4, 1b
+
+        ret
+endfunc
+
 func ff_scalarproduct_fixed_rvv, zve64x
         li      t1, 1 << 30
         vsetvli t0, zero, e64, m8, ta, ma
-- 
2.42.0