Errata for The Software Vectorization Handbook

First Printing

• page 25: replace "a2-b3" with "a2-b2" in destination (courtesy Naftali Schwartz).
• page 27: replace "stores the 16-bit sums" with "stores the 32-bit sums".
• page 35: replace the immediate 230 with 245 in the shufps that is equal to movshdup.
• page 38: change theoretical peak into 12GFLOPS for single- and 6GFLOPS for double-precision FP computations.
• page 94: add "T" to two occurrences of "pcmpeq" in expansion table (viz. "pcmpeqT") and "d" to "pcmpeq" in example at the bottom (viz. "pcmpeqd").
• page 94: replace "constants -1 and -17" with "constants -1 and 17" (courtesy Naftali Schwartz).
• page 112: replace "pcmpneb,w,d" entries in Table 5.8 with "emulate with pcmpeqb,w,d" (no direct support of "ne" instructions).
• page 113: remove pcmpneb,w row completely from table (no direct support of "ne" instructions).
• page 127: remove ")" from caption of Figure 6.2.
• page 190: remove "is" from "vector implementation is shown above".
• page 209: remove "LOOP" from "BLOCK LOOP WAS VECTORIZED" in Table 9.5.
• page 215: add brackets "(" ")" around switches "-O3 -Qipo -Qprof_use -QxP".

Second Printing

• none known yet