Protection from attempting to do SIMD rans encoding on small data

We can't just change do_simd as we may then have a combination of SIMD
(pack or rle-meta) and non-SIMD (rle-meta or rle-data).  So instead we
remove the ability to do SIMD and call again.

This only happens when we have large blocks of data that enable SIMD
mode which then turn out to be heavily compressible after PACK or RLE
to yield a small sub-stream which isn't SIMD compressible.

Author: jkbonfield

htscodecs/rANS_static4x16pr.c

@@ -1169,6 +1169,10 @@ unsigned char *rans_compress_to_4x16(unsigned char *in, unsigned int in_size,
         return NULL;
     }
 
+    int orig_order = order;
+    int orig_in_size = in_size;
+    unsigned char *orig_in = in;
+
     unsigned int c_meta_len;
     uint8_t *meta = NULL, *rle = NULL, *packed = NULL;
     uint8_t *out_free = NULL;
@@ -1385,6 +1389,14 @@ unsigned char *rans_compress_to_4x16(unsigned char *in, unsigned int in_size,
             int sz = var_put_u32(out+c_meta_len, out_end, in_size);
             c_meta_len += sz;
             *out_size -= sz;
+
+            if (do_simd && in_size < 32) {
+                free(packed);
+                return rans_compress_to_4x16(orig_in, orig_in_size,
+                                             out, out_size,
+                                             orig_order & ~(RANS_ORDER_X32 |
+                                                            RANS_ORDER_SIMD_AUTO));
+            }
         }
     } else if (do_pack) {
         out[0] &= ~RANS_ORDER_PACK;
@@ -1430,6 +1442,15 @@ unsigned char *rans_compress_to_4x16(unsigned char *in, unsigned int in_size,
                 return NULL;
             }
             c_rmeta_len = *out_size - (c_meta_len+sz+5);
+            if (do_simd && (rmeta_len < 32 || rle_len < 32)) {
+                free(packed);
+                free(rle);
+                free(meta);
+                return rans_compress_to_4x16(orig_in, orig_in_size,
+                                             out, out_size,
+                                             orig_order & ~(RANS_ORDER_X32 |
+                                                            RANS_ORDER_SIMD_AUTO));
+            }
             if (!rans_enc_func(do_simd, 0)(meta, rmeta_len, out+c_meta_len+sz+5, &c_rmeta_len)) {
                 free(out_free);
                 free(rle);
@@ -1496,13 +1517,31 @@ unsigned char *rans_compress_to_4x16(unsigned char *in, unsigned int in_size,
         if (in_size)
             memcpy(out+c_meta_len, in, in_size);
         *out_size = in_size;
-    }
+   }
 
     free(rle);
     free(packed);
 
     *out_size += c_meta_len;
 
+// Validation mode
+#ifdef VALIDATE_RANS
+    unsigned int decoded_size = orig_in_size;
+    unsigned char *decoded = malloc(decoded_size);
+    decoded = rans_uncompress_to_4x16(out, *out_size,
+                                      decoded, &decoded_size);
+    if (!decoded ||
+        decoded_size != orig_in_size ||
+        memcmp(orig_in, decoded, orig_in_size) != 0) {
+        fprintf(stderr, "rans round trip failed for order %d. Written to fd 5\n", orig_order);
+        if (write(5, orig_in, orig_in_size) < 0)
+            abort();
+        abort();
+    }
+    free(decoded);
+#endif
+
+
     return out;
 }