From ca664f2254a96e9fc9c633fea5eac11f51d93923 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= <remi@remlab.net>
Date: Wed, 15 Nov 2023 21:31:17 +0200
Subject: [PATCH] lavc/flacdsp: R-V V LPC16 function

In this case, the inner loop computing the scalar product can be reduced
to just one multiplication and one sum even with 128-bit vectors. The
result is a lot simpler, but also brings more modest performance gains:

flac_lpc_16_13_c:       15241.0
flac_lpc_16_13_rvv_i32: 11230.0
flac_lpc_16_16_c:       17884.0
flac_lpc_16_16_rvv_i32: 12125.7
flac_lpc_16_29_c:       27847.7
flac_lpc_16_29_rvv_i32: 10494.0
flac_lpc_16_32_c:       30051.5
flac_lpc_16_32_rvv_i32: 10355.0
---
 libavcodec/riscv/flacdsp_init.c | 17 ++++++++++++-----
 libavcodec/riscv/flacdsp_rvv.S  | 23 +++++++++++++++++++++++
 2 files changed, 35 insertions(+), 5 deletions(-)

diff --git a/libavcodec/riscv/flacdsp_init.c b/libavcodec/riscv/flacdsp_init.c
index 0cdbf2fec2..6cfb50ead8 100644
--- a/libavcodec/riscv/flacdsp_init.c
+++ b/libavcodec/riscv/flacdsp_init.c
@@ -25,6 +25,8 @@
 #include "libavutil/riscv/cpu.h"
 #include "libavcodec/flacdsp.h"
 
+void ff_flac_lpc16_rvv(int32_t *decoded, const int coeffs[32],
+                       int pred_order, int qlevel, int len);
 void ff_flac_lpc32_rvv(int32_t *decoded, const int coeffs[32],
                        int pred_order, int qlevel, int len);
 void ff_flac_lpc32_rvv_simple(int32_t *decoded, const int coeffs[32],
@@ -61,16 +63,20 @@ void ff_flac_decorrelate_ms_32_rvv(uint8_t **out, int32_t **in,
 av_cold void ff_flacdsp_init_riscv(FLACDSPContext *c, enum AVSampleFormat fmt,
                                    int channels)
 {
-#if HAVE_RVV && (__riscv_xlen >= 64)
+#if HAVE_RVV
     int flags = av_get_cpu_flags();
 
     if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
         int vlenb = ff_get_rv_vlenb();
 
-        if (vlenb == 16)
-            c->lpc32 = ff_flac_lpc32_rvv;
-        else if (vlenb > 16)
-            c->lpc32 = ff_flac_lpc32_rvv_simple;
+        if (vlenb >= 16) {
+            c->lpc16 = ff_flac_lpc16_rvv;
+# if (__riscv_xlen >= 64)
+            if (vlenb > 16)
+                c->lpc32 = ff_flac_lpc32_rvv_simple;
+            else
+                c->lpc32 = ff_flac_lpc32_rvv;
+        }
 
         switch (fmt) {
         case AV_SAMPLE_FMT_S16:
@@ -111,6 +117,7 @@ av_cold void ff_flacdsp_init_riscv(FLACDSPContext *c, enum AVSampleFormat fmt,
             c->decorrelate[2] = ff_flac_decorrelate_rs_32_rvv;
             c->decorrelate[3] = ff_flac_decorrelate_ms_32_rvv;
             break;
+# endif
         }
     }
 #endif
diff --git a/libavcodec/riscv/flacdsp_rvv.S b/libavcodec/riscv/flacdsp_rvv.S
index b1724f5500..2a0b50f7a9 100644
--- a/libavcodec/riscv/flacdsp_rvv.S
+++ b/libavcodec/riscv/flacdsp_rvv.S
@@ -20,6 +20,29 @@
 
 #include "libavutil/riscv/asm.S"
 
+func ff_flac_lpc16_rvv, zve32x
+        vsetvli zero, a2, e32, m8, ta, ma
+        vle32.v v8, (a1)
+        sub     a4, a4, a2
+        vle32.v v16, (a0)
+        sh2add  a0, a2, a0
+        vmv.s.x v0, zero
+1:
+        vmul.vv v24, v8, v16
+        lw      t0, (a0)
+        vredsum.vs v24, v24, v0
+        addi    a4, a4, -1
+        vmv.x.s t1, v24
+        sra     t1, t1, a3
+        add     t0, t0, t1
+        vslide1down.vx v16, v16, t0
+        sw      t0, (a0)
+        addi    a0, a0, 4
+        bnez    a4, 1b
+
+        ret
+endfunc
+
 #if (__riscv_xlen == 64)
 func ff_flac_lpc32_rvv, zve32x
         addi    t2, a2, -16