[FFmpeg-devel] [PATCH] lavc/aacencdsp: R-V V quant_bands

[Rémi Denis-Courmont]

T-Head C908:
quant_bands_signed_c:        576.0
quant_bands_signed_rvv_f32:   48.7
quant_bands_unsigned_c:      414.2
quant_bands_unsigned_rvv_f32: 31.7

SpacemiT X60:
quant_bands_signed_c:        497.7
quant_bands_signed_rvv_f32:   23.0
quant_bands_unsigned_c:      353.5
quant_bands_unsigned_rvv_f32: 16.2
---
 libavcodec/riscv/aacencdsp_init.c |  3 +++
 libavcodec/riscv/aacencdsp_rvv.S  | 31 +++++++++++++++++++++++++++++++
 2 files changed, 34 insertions(+)

diff --git a/libavcodec/riscv/aacencdsp_init.c b/libavcodec/riscv/aacencdsp_init.c
index b27af9d973..73bc8d8fa3 100644
--- a/libavcodec/riscv/aacencdsp_init.c
+++ b/libavcodec/riscv/aacencdsp_init.c
@@ -26,6 +26,8 @@
 #include "libavcodec/aacencdsp.h"
 
 void ff_abs_pow34_rvv(float *out, const float *in, const int size);
+void ff_aac_quant_bands_rvv(int *, const float *, const float *, int, int,
+                            int, const float, const float);
 
 av_cold void ff_aacenc_dsp_init_riscv(AACEncDSPContext *s)
 {
@@ -35,6 +37,7 @@ av_cold void ff_aacenc_dsp_init_riscv(AACEncDSPContext *s)
     if (flags & AV_CPU_FLAG_RVV_F32) {
         if (flags & AV_CPU_FLAG_RVB_ADDR) {
             s->abs_pow34 = ff_abs_pow34_rvv;
+            s->quant_bands = ff_aac_quant_bands_rvv;
         }
     }
 #endif
diff --git a/libavcodec/riscv/aacencdsp_rvv.S b/libavcodec/riscv/aacencdsp_rvv.S
index 4c7a874d77..ddb4155398 100644
--- a/libavcodec/riscv/aacencdsp_rvv.S
+++ b/libavcodec/riscv/aacencdsp_rvv.S
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2023 Institue of Software Chinese Academy of Sciences (ISCAS).
+ * Copyright © 2024 Rémi Denis-Courmont.
  *
  * This file is part of FFmpeg.
  *
@@ -36,3 +37,33 @@ func ff_abs_pow34_rvv, zve32f
 
         ret
 endfunc
+
+func ff_aac_quant_bands_rvv, zve32f
+NOHWF   fmv.w.x     fa0, a6
+NOHWF   fmv.w.x     fa1, a7
+        fcvt.s.w    ft0, a5
+        bnez        a4, .L1 # signed?
+        .irp signed, 0, 1
+.L\signed:
+        vsetvli     t0, a3, e32, m8, ta, ma
+        vle32.v     v16, (a2)
+        sub         a3, a3, t0
+        .if \signed
+        vle32.v     v8, (a1)
+        sh2add      a1, t0, a1
+        .endif
+        vfmul.vf    v16, v16, fa0
+        sh2add      a2, t0, a2
+        vfadd.vf    v16, v16, fa1
+        vfmin.vf    v16, v16, ft0
+        .if \signed
+        vfsgnjx.vv  v16, v16, v8
+        .endif
+        vfcvt.x.f.v v16, v16
+        vse32.v     v16, (a0)
+        sh2add      a0, t0, a0
+        bnez        a3, .L\signed
+
+        ret
+        .endr
+endfunc
-- 
2.45.1