ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Common/Hash: use __crc32d() intrinsic on ARM64

This commit is contained in:
Tillmann Karras 2017-05-03 06:40:47 +01:00
parent e4e37b3a82
commit 25f61f0329
1 changed files with 12 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
Source/Core/Common/Hash.cpp
View File

@ -9,6 +9,10 @@
#include "Common/CommonFuncs.h" #include "Common/CommonFuncs.h"
#include "Common/Intrinsics.h" #include "Common/Intrinsics.h"
#ifdef _M_ARM_64
#include <arm_acle.h>
#endif
static u64 (*ptrHashFunction)(const u8* src, u32 len, u32 samples) = &GetMurmurHash3; static u64 (*ptrHashFunction)(const u8* src, u32 len, u32 samples) = &GetMurmurHash3;
// uint32_t // uint32_t
@ -293,47 +297,26 @@ u64 GetCRC32(const u8* src, u32 len, u32 samples)
if (Step < 1) if (Step < 1)
Step = 1; Step = 1;
// We should be able to use intrinsics for this
// Too bad the intrinsics for this instruction was added in GCC 4.9.1
// The Android NDK (as of r10e) only has GCC 4.9
// Once the Android NDK has a newer GCC version, update these to use intrinsics
while (data < end - Step * 3) while (data < end - Step * 3)
{ {
asm("crc32x %w[res], %w[two], %x[three]" h[0] = __crc32d(h[0], data[Step * 0]);
: [res] "=r"(h[0]) h[1] = __crc32d(h[1], data[Step * 1]);
: [two] "r"(h[0]), [three] "r"(data[Step * 0])); h[2] = __crc32d(h[2], data[Step * 2]);
asm("crc32x %w[res], %w[two], %x[three]" h[3] = __crc32d(h[3], data[Step * 3]);
: [res] "=r"(h[1])
: [two] "r"(h[1]), [three] "r"(data[Step * 1]));
asm("crc32x %w[res], %w[two], %x[three]"
: [res] "=r"(h[2])
: [two] "r"(h[2]), [three] "r"(data[Step * 2]));
asm("crc32x %w[res], %w[two], %x[three]"
: [res] "=r"(h[3])
: [two] "r"(h[3]), [three] "r"(data[Step * 3]));
data += Step * 4; data += Step * 4;
} }
if (data < end - Step * 0) if (data < end - Step * 0)
asm("crc32x %w[res], %w[two], %x[three]" h[0] = __crc32d(h[0], data[Step * 0]);
: [res] "=r"(h[0])
: [two] "r"(h[0]), [three] "r"(data[Step * 0]));
if (data < end - Step * 1) if (data < end - Step * 1)
asm("crc32x %w[res], %w[two], %x[three]" h[1] = __crc32d(h[1], data[Step * 1]);
: [res] "=r"(h[1])
: [two] "r"(h[1]), [three] "r"(data[Step * 1]));
if (data < end - Step * 2) if (data < end - Step * 2)
asm("crc32x %w[res], %w[two], %x[three]" h[2] = __crc32d(h[2], data[Step * 2]);
: [res] "=r"(h[2])
: [two] "r"(h[2]), [three] "r"(data[Step * 2]));
if (len & 7) if (len & 7)
{ {
u64 temp = 0; u64 temp = 0;
memcpy(&temp, end, len & 7); memcpy(&temp, end, len & 7);
asm("crc32x %w[res], %w[two], %x[three]" h[0] = __crc32d(h[0], temp);
: [res] "=r"(h[0])
: [two] "r"(h[0]), [three] "r"(temp));
} }
// FIXME: is there a better way to combine these partial hashes? // FIXME: is there a better way to combine these partial hashes?