From 88d973a5d658dc61dfd32e87b062724def46addc Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= <remi@remlab.net>
Date: Sun, 12 May 2024 20:02:03 +0300
Subject: [PATCH] lavc/flacdsp: R-V V flac_wasted33

T-Head C908:
flac_wasted_33_c:       786.2
flac_wasted_33_rvv_i64: 486.5
---
 libavcodec/riscv/flacdsp_init.c |  4 ++++
 libavcodec/riscv/flacdsp_rvv.S  | 32 ++++++++++++++++++++++++++++++++
 2 files changed, 36 insertions(+)

diff --git a/libavcodec/riscv/flacdsp_init.c b/libavcodec/riscv/flacdsp_init.c
index 454787470b..4f1652dbe7 100644
--- a/libavcodec/riscv/flacdsp_init.c
+++ b/libavcodec/riscv/flacdsp_init.c
@@ -32,6 +32,7 @@ void ff_flac_lpc32_rvv(int32_t *decoded, const int coeffs[32],
 void ff_flac_lpc32_rvv_simple(int32_t *decoded, const int coeffs[32],
                               int pred_order, int qlevel, int len);
 void ff_flac_wasted32_rvv(int32_t *, int shift, int len);
+void ff_flac_wasted33_rvv(int64_t *, const int32_t *, int shift, int len);
 void ff_flac_decorrelate_indep2_16_rvv(uint8_t **out, int32_t **in,
                                        int channels, int len, int shift);
 void ff_flac_decorrelate_indep4_16_rvv(uint8_t **out, int32_t **in,
@@ -84,6 +85,9 @@ av_cold void ff_flacdsp_init_riscv(FLACDSPContext *c, enum AVSampleFormat fmt,
 
         c->wasted32 = ff_flac_wasted32_rvv;
 
+        if (flags & AV_CPU_FLAG_RVV_I64)
+            c->wasted33 = ff_flac_wasted33_rvv;
+
 # if (__riscv_xlen >= 64)
         switch (fmt) {
         case AV_SAMPLE_FMT_S16:
diff --git a/libavcodec/riscv/flacdsp_rvv.S b/libavcodec/riscv/flacdsp_rvv.S
index d7009cdec2..6287faa260 100644
--- a/libavcodec/riscv/flacdsp_rvv.S
+++ b/libavcodec/riscv/flacdsp_rvv.S
@@ -115,6 +115,38 @@ func ff_flac_wasted32_rvv, zve32x
         ret
 endfunc
 
+func ff_flac_wasted33_rvv, zve64x
+        srli         t0, a2, 5
+        li           t1, 1
+        bnez         t0, 2f
+        sll          a2, t1, a2
+1:
+        vsetvli      t0, a3, e32, m4, ta, ma
+        vle32.v      v8, (a1)
+        sub          a3, a3, t0
+        vwmulsu.vx   v16, v8, a2
+        sh2add       a1, t0, a1
+        vse64.v      v16, (a0)
+        sh3add       a0, t0, a0
+        bnez         a3, 1b
+
+        ret
+
+2:      // Pessimistic case: wasted >= 32
+        vsetvli      t0, a3, e32, m4, ta, ma
+        vle32.v      v8, (a1)
+        sub          a3, a3, t0
+        vwcvtu.x.x.v v16, v8
+        sh2add       a1, t0, a1
+        vsetvli      zero, zero, e64, m8, ta, ma
+        vsll.vx      v16, v16, a2
+        vse64.v      v16, (a0)
+        sh3add       a0, t0, a0
+        bnez         a3, 2b
+
+        ret
+endfunc
+
 #if (__riscv_xlen == 64)
 func ff_flac_decorrelate_indep2_16_rvv, zve32x
         ld      a0,  (a0)