github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

Remove unneeded code at preproc stage

Browse Source 
If we know we have BMI2 we shouldn't produce the fallback code.
This commit is contained in:
Matthew Barr 2015-10-30 10:43:43 +11:00
parent aafbd96a1d
commit 1f47b82106
2 changed files with 12 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/util/bitutils.h
View File

@ -254,7 +254,7 @@ u32 compress32(u32 x, u32 m) {
#if defined(__BMI2__)
// BMI2 has a single instruction for this operation.
return _pext_u32(x, m);
#endif
#else
// Return zero quickly on trivial cases
if ((x & m) == 0) {
@ -281,6 +281,7 @@ u32 compress32(u32 x, u32 m) {
}
return x;
#endif
}
static really_inline
@ -288,7 +289,7 @@ u64a compress64(u64a x, u64a m) {
#if defined(ARCH_X86_64) && defined(__BMI2__)
// BMI2 has a single instruction for this operation.
return _pext_u64(x, m);
#endif
#else
// Return zero quickly on trivial cases
if ((x & m) == 0) {
@ -316,6 +317,7 @@ u64a compress64(u64a x, u64a m) {
}
return x;
#endif
}
static really_inline
@ -323,7 +325,7 @@ u32 expand32(u32 x, u32 m) {
#if defined(__BMI2__)
// BMI2 has a single instruction for this operation.
return _pdep_u32(x, m);
#endif
#else
// Return zero quickly on trivial cases
if (!x || !m) {
@ -355,6 +357,7 @@ u32 expand32(u32 x, u32 m) {
}
return x & m0; // clear out extraneous bits
#endif
}
static really_inline
@ -362,7 +365,7 @@ u64a expand64(u64a x, u64a m) {
#if defined(ARCH_X86_64) && defined(__BMI2__)
// BMI2 has a single instruction for this operation.
return _pdep_u64(x, m);
#endif
#else
// Return zero quickly on trivial cases
if (!x || !m) {
@ -395,6 +398,7 @@ u64a expand64(u64a x, u64a m) {
}
return x & m0; // clear out extraneous bits
#endif
}

6
src/util/shuffle.h
View File

@ -51,7 +51,7 @@ u32 shuffleDynamic32(u32 x, u32 mask) {
#if defined(HAVE_PEXT)
// Intel BMI2 can do this operation in one instruction.
return _pext_u32(x, mask);
#endif
#else
u32 result = 0, num = 1;
while (mask != 0) {
@ -63,6 +63,7 @@ u32 shuffleDynamic32(u32 x, u32 mask) {
num <<= 1;
}
return result;
#endif
}
static really_inline
@ -70,7 +71,7 @@ u32 shuffleDynamic64(u64a x, u64a mask) {
#if defined(HAVE_PEXT) && defined(ARCH_64_BIT)
// Intel BMI2 can do this operation in one instruction.
return _pext_u64(x, mask);
#endif
#else
u32 result = 0, num = 1;
while (mask != 0) {
@ -82,6 +83,7 @@ u32 shuffleDynamic64(u64a x, u64a mask) {
num <<= 1;
}
return result;
#endif
}
#undef HAVE_PEXT