Merge pull request #2689 from danlark1/dev

Optimize zstd decompression by another x%
2021-06-29 11:34:36 -07:00
parent 88f107b5b9 2c2c9e7dfd
commit 094b26081f
2 changed files with 103 additions and 106 deletions
@@ -332,7 +332,16 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 c
    U32 const regMask = sizeof(bitContainer)*8 - 1;
    /* if start > regMask, bitstream is corrupted, and result is undefined */
    assert(nbBits < BIT_MASK_SIZE);
+    /* x86 transform & ((1 << nbBits) - 1) to bzhi instruction, it is better
+     * than accessing memory. When bmi2 instruction is not present, we consider
+     * such cpus old (pre-Haswell, 2013) and their performance is not of that
+     * importance.
+     */
+#if defined(__x86_64__) || defined(_M_X86)
+    return (bitContainer >> (start & regMask)) & ((((U64)1) << nbBits) - 1);
+#else
    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
+#endif
 }

 MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)