Document the fact that the AltiVec implementation uses the same modified algorithms as the SSE2 implementation

git-svn-id: svn+ssh://svn.code.sf.net/p/libjpeg-turbo/code/trunk@1473 632fc199-4ca6-4c93-a231-07263d6284db
2014-12-23 02:42:59 +00:00
parent 45453085e7
commit ff30c63934
3 changed files with 85 additions and 1 deletions
--- a/simd/jidctfst-altivec.c
+++ b/simd/jidctfst-altivec.c
@@ -77,6 +77,16 @@
  \
  tmp7 = vec_add(z11, z13);  \
  \
+  /* To avoid overflow...  \
+   *  \
+   * (Original)  \
+   * tmp12 = -2.613125930 * z10 + z5;  \
+   *  \
+   * (This implementation)  \
+   * tmp12 = (-1.613125930 - 1) * z10 + z5;  \
+   *       = -1.613125930 * z10 - z10 + z5;  \
+   */  \
+  \
  z5 = vec_add(z10s, z12s);  \
  z5 = vec_madds(z5, pw_F1847, zero);  \
  \