summaryrefslogtreecommitdiff
path: root/lib/zstd/common/bitstream.h
diff options
context:
space:
mode:
authorDmitry Torokhov <dmitry.torokhov@gmail.com>2023-02-21 11:19:49 -0800
committerDmitry Torokhov <dmitry.torokhov@gmail.com>2023-02-21 11:19:49 -0800
commit7ae9fb1b7ecbb5d85d07857943f677fd1a559b18 (patch)
treedbdd35328f43569c38c4ce193cefd7d2b6b9fbfd /lib/zstd/common/bitstream.h
parent9c445d2637c938a800fcc8b5f0b10e60c94460c7 (diff)
parent9e69e845ae95227949c400af1037dca023f73038 (diff)
Merge branch 'next' into for-linus
Prepare input updates for 6.3 merge window.
Diffstat (limited to 'lib/zstd/common/bitstream.h')
-rw-r--r--lib/zstd/common/bitstream.h9
1 files changed, 9 insertions, 0 deletions
diff --git a/lib/zstd/common/bitstream.h b/lib/zstd/common/bitstream.h
index 28248abe8612..feef3a1b1d60 100644
--- a/lib/zstd/common/bitstream.h
+++ b/lib/zstd/common/bitstream.h
@@ -313,7 +313,16 @@ 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)