mirrors/xmrig
1
0
Fork 1
mirror of https://github.com/xmrig/xmrig.git synced 2025-12-06 23:52:38 -05:00
Code Issues Projects Releases Wiki Activity

Merge pull request #1557 from SChernykh/dev

Browse Source 
Refactored fma_soft() in randomx_vm.cl
This commit is contained in:
xmrig
2020-02-19 00:28:17 +07:00
committed by GitHub
parent 2bc5fb10a7 887c891ab2
commit 88c7aca6f5
2 changed files with 1380 additions and 1376 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2728
src/backend/opencl/cl/rx/randomx_cl.h
View File

File diff suppressed because it is too large Load Diff

28
src/backend/opencl/cl/rx/randomx_vm.cl
View File

@@ -1430,9 +1430,13 @@ double fma_soft(double a, double b, double c, uint32_t rounding_mode)
const uint64_t exponent_size = 11;
const uint64_t exponent_mask = (1 << exponent_size) - 1;
const uint32_t exponent_a = (as_uint2(a).y >> 20) & exponent_mask;
const uint32_t exponent_b = (as_uint2(b).y >> 20) & exponent_mask;
const uint32_t exponent_c = (as_uint2(c).y >> 20) & exponent_mask;
uint2 a2 = as_uint2(a);
uint2 b2 = as_uint2(b);
uint2 c2 = as_uint2(c);
const uint32_t exponent_a = (a2.y >> 20) & exponent_mask;
const uint32_t exponent_b = (b2.y >> 20) & exponent_mask;
const uint32_t exponent_c = (c2.y >> 20) & exponent_mask;
if ((exponent_a == 2047) || (exponent_b == 2047) || (exponent_c == 2047))
{
@@ -1440,17 +1444,17 @@ double fma_soft(double a, double b, double c, uint32_t rounding_mode)
return as_double(inf);
}
uint64_t mantissa_a = (as_ulong(a) & mantissa_mask);
uint64_t mantissa_b = (as_ulong(b) & mantissa_mask);
uint64_t mantissa_c = (as_ulong(c) & mantissa_mask);
const uint32_t sign_a = a2.y >> 31;
const uint32_t sign_b = b2.y >> 31;
const uint32_t sign_c = c2.y >> 31;
((uint2*)&mantissa_a)->y |= 1U << 20;
((uint2*)&mantissa_b)->y |= 1U << 20;
((uint2*)&mantissa_c)->y |= 1U << 20;
a2.y = (a2.y & ((1U << 20) - 1)) | (1U << 20);
b2.y = (b2.y & ((1U << 20) - 1)) | (1U << 20);
c2.y = (c2.y & ((1U << 20) - 1)) | (1U << 20);
const uint32_t sign_a = as_uint2(a).y >> 31;
const uint32_t sign_b = as_uint2(b).y >> 31;
const uint32_t sign_c = as_uint2(c).y >> 31;
uint64_t mantissa_a = as_ulong(a2);
uint64_t mantissa_b = as_ulong(b2);
uint64_t mantissa_c = as_ulong(c2);
uint64_t mul_result[2];
mul_result[0] = mantissa_a * mantissa_b;