From 0a61ee08b0b8c0e820ef6966c23b3eea58abbfaa Mon Sep 17 00:00:00 2001
From: Jean-Marc Valin <jmvalin@amazon.com>
Date: Tue, 1 Aug 2023 19:14:29 -0400
Subject: Bring back support for non-multiples of 8

---
 dnn/vec.h      | 11 +++++++++--
 dnn/vec_avx.h  | 11 +++++++++--
 dnn/vec_neon.h | 11 +++++++++--
 3 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/dnn/vec.h b/dnn/vec.h
index 30617de8..5b6951bb 100644
--- a/dnn/vec.h
+++ b/dnn/vec.h
@@ -120,9 +120,16 @@ static inline void sgemv16x1(float *out, const float *weights, int rows, int col
 
 static inline void sgemv(float *out, const float *weights, int rows, int cols, int col_stride, const float *x)
 {
-   celt_assert((rows&7) == 0);
    if ((rows&0xf) == 0) sgemv16x1(out, weights, rows, cols, col_stride, x);
-   else sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else if ((rows&0x7) == 0) sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else {
+      int i, j;
+      for (i=0;i<rows;i++)
+      {
+         out[i] = 0;
+         for (j=0;j<cols;j++) out[i] += weights[j*col_stride + i]*x[j];
+      }
+   }
 }
 
 static inline void sparse_sgemv8x4(float *out, const float *w, const int *idx, int rows, const float *x)
diff --git a/dnn/vec_avx.h b/dnn/vec_avx.h
index b31caa88..4747bb41 100644
--- a/dnn/vec_avx.h
+++ b/dnn/vec_avx.h
@@ -725,9 +725,16 @@ static inline void sgemv8x1(float *out, const float *weights, int rows, int cols
 
 static inline void sgemv(float *out, const float *weights, int rows, int cols, int col_stride, const float *x)
 {
-   celt_assert((rows&7) == 0);
    if ((rows&0xf) == 0) sgemv16x1(out, weights, rows, cols, col_stride, x);
-   else sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else if ((rows&0x7) == 0) sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else {
+      int i, j;
+      for (i=0;i<rows;i++)
+      {
+         out[i] = 0;
+         for (j=0;j<cols;j++) out[i] += weights[j*col_stride + i]*x[j];
+      }
+   }
 }
 
 static inline void sparse_sgemv8x4(float *out, const float *weights, const int *idx, int rows, const float *x)
diff --git a/dnn/vec_neon.h b/dnn/vec_neon.h
index af3b204f..48e3eaa1 100644
--- a/dnn/vec_neon.h
+++ b/dnn/vec_neon.h
@@ -276,9 +276,16 @@ static inline void sgemv8x1(float *out, const float *weights, int rows, int cols
 
 static inline void sgemv(float *out, const float *weights, int rows, int cols, int col_stride, const float *x)
 {
-   celt_assert((rows&7) == 0);
    if ((rows&0xf) == 0) sgemv16x1(out, weights, rows, cols, col_stride, x);
-   else sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else if ((rows&0x7) == 0) sgemv8x1(out, weights, rows, cols, col_stride, x);
+   else {
+      int i, j;
+      for (i=0;i<rows;i++)
+      {
+         out[i] = 0;
+         for (j=0;j<cols;j++) out[i] += weights[j*col_stride + i]*x[j];
+      }
+   }
 }
 
 /* Temporarily use unoptimized version */
-- 
cgit v1.2.3