From e33ce0d9dd8136d9d1a5eb17863c2f6d4d5f131f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= <remi@remlab.net>
Date: Wed, 4 Oct 2023 22:50:47 +0300
Subject: [PATCH] lavu/fixed_dsp: R-V V fmul_window_scaled

vector_fmul_window_scaled_fixed_c:       4393.7
vector_fmul_window_scaled_fixed_rvv_i64: 1642.7
---
 libavutil/riscv/fixed_dsp_init.c |  7 ++++-
 libavutil/riscv/fixed_dsp_rvv.S  | 48 ++++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 1 deletion(-)

diff --git a/libavutil/riscv/fixed_dsp_init.c b/libavutil/riscv/fixed_dsp_init.c
index 6469b45374..cd318af486 100644
--- a/libavutil/riscv/fixed_dsp_init.c
+++ b/libavutil/riscv/fixed_dsp_init.c
@@ -25,6 +25,9 @@
 #include "libavutil/cpu.h"
 #include "libavutil/fixed_dsp.h"
 
+void ff_vector_fmul_window_scaled_rvv(int16_t *dst, const int32_t *src0,
+                                      const int32_t *src1, const int32_t *win,
+                                      int len, uint8_t bits);
 void ff_vector_fmul_window_fixed_rvv(int32_t *dst, const int32_t *src0,
                                      const int32_t *src1, const int32_t *win,
                                      int len);
@@ -43,8 +46,10 @@ 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)) {
-        if (flags & AV_CPU_FLAG_RVV_I64)
+        if (flags & AV_CPU_FLAG_RVV_I64) {
+            fdsp->vector_fmul_window_scaled = ff_vector_fmul_window_scaled_rvv;
             fdsp->vector_fmul_window = ff_vector_fmul_window_fixed_rvv;
+        }
 
         fdsp->vector_fmul = ff_vector_fmul_fixed_rvv;
         fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_fixed_rvv;
diff --git a/libavutil/riscv/fixed_dsp_rvv.S b/libavutil/riscv/fixed_dsp_rvv.S
index 68de6d7e1b..6bac5813b8 100644
--- a/libavutil/riscv/fixed_dsp_rvv.S
+++ b/libavutil/riscv/fixed_dsp_rvv.S
@@ -20,6 +20,54 @@
 
 #include "asm.S"
 
+func ff_vector_fmul_window_scaled_rvv, zve64x
+        csrwi   vxrm, 0
+        vsetvli t0, zero, e16, m1, ta, ma
+        sh2add  a2, a4, a2
+        vid.v   v0
+        sh3add  t3, a4, a3
+        vadd.vi v0, v0, 1
+        sh2add  t0, a4, a0
+1:
+        vsetvli t2, a4, e16, m1, ta, ma
+        slli    t4, t2, 2
+        slli    t1, t2, 1
+        vrsub.vx v2, v0, t2
+        sub     t3, t3, t4
+        vsetvli zero, zero, e32, m2, ta, ma
+        sub     a2, a2, t4
+        vle32.v v8, (t3)
+        sub     t0, t0, t1
+        vle32.v v4, (a2)
+        sub     a4, a4, t2
+        vrgatherei16.vv v28, v8, v2
+        vle32.v v16, (a1)
+        add     a1, a1, t4
+        vrgatherei16.vv v20, v4, v2
+        vle32.v v24, (a3)
+        add     a3, a3, t4
+        vwmul.vv v12, v16, v28
+        vwmul.vv v8, v16, v24
+        // vwnmsac.vv does _not_ exist so multiply & subtract separately
+        vwmul.vv v4, v20, v24
+        vwmacc.vv v8, v20, v28
+        vsetvli zero, zero, e64, m4, ta, ma
+        vsub.vv v12, v12, v4
+        vsetvli zero, zero, e32, m2, ta, ma
+        vnclip.wi v16, v8, 31
+        vnclip.wi v20, v12, 31
+        vsetvli zero, zero, e16, m1, ta, ma
+        vnclip.wx v8, v16, a5
+        vnclip.wx v12, v20, a5
+        vrgatherei16.vv v16, v8, v2
+        vse16.v v12, (a0)
+        add     a0, a0, t1
+        vse16.v v16, (t0)
+        bnez    a4, 1b
+
+        ret
+endfunc
+
 func ff_vector_fmul_window_fixed_rvv, zve64x
         csrwi   vxrm, 0
         vsetvli t0, zero, e16, m1, ta, ma