griefith/test
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cleanup: rename sse2 specific functions to avoid accidentally using them

Browse Source 
Ref #115098

Pull Request: https://projects.blender.org/blender/blender/pulls/115098
This commit is contained in:
Brecht Van Lommel
2023-11-22 19:41:47 +01:00
committed by Gitea
parent 6ec0e0104a
commit b9227da2db
1 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
intern/cycles/util/color.h
View File

@@ -253,7 +253,7 @@ ccl_device float3 xyY_to_xyz(float x, float y, float Y)
* exp = exponent, encoded as uint32_t
* e2coeff = 2^(127/exponent - 127) * bias_coeff^(1/exponent), encoded as uint32_t
*/
template<unsigned exp, unsigned e2coeff> ccl_device_inline float4 fastpow(const float4 &arg)
template<unsigned exp, unsigned e2coeff> ccl_device_inline float4 fastpow_sse2(const float4 &arg)
{
float4 ret = arg * cast(make_int4(e2coeff));
ret = make_float4(cast(ret));
@@ -263,7 +263,7 @@ template<unsigned exp, unsigned e2coeff> ccl_device_inline float4 fastpow(const
}
/* Improve x ^ 1.0f/5.0f solution with Newton-Raphson method */
ccl_device_inline float4 improve_5throot_solution(const float4 &old_result, const float4 &x)
ccl_device_inline float4 improve_5throot_solution_sse2(const float4 &old_result, const float4 &x)
{
float4 approx2 = old_result * old_result;
float4 approx4 = approx2 * approx2;
@@ -273,7 +273,7 @@ ccl_device_inline float4 improve_5throot_solution(const float4 &old_result, cons
}
/* Calculate powf(x, 2.4). Working domain: 1e-10 < x < 1e+10 */
ccl_device_inline float4 fastpow24(const float4 &arg)
ccl_device_inline float4 fastpow24_sse2(const float4 &arg)
{
/* max, avg and |avg| errors were calculated in gcc without FMA instructions
* The final precision should be better than powf in glibc */
@@ -281,27 +281,27 @@ ccl_device_inline float4 fastpow24(const float4 &arg)
/* Calculate x^4/5, coefficient 0.994 was constructed manually to minimize avg error */
/* 0x3F4CCCCD = 4/5 */
/* 0x4F55A7FB = 2^(127/(4/5) - 127) * 0.994^(1/(4/5)) */
float4 x = fastpow<0x3F4CCCCD, 0x4F55A7FB>(
float4 x = fastpow_sse2<0x3F4CCCCD, 0x4F55A7FB>(
arg); // error max = 0.17 avg = 0.0018 |avg| = 0.05
float4 arg2 = arg * arg;
float4 arg4 = arg2 * arg2;
/* error max = 0.018 avg = 0.0031 |avg| = 0.0031 */
x = improve_5throot_solution(x, arg4);
x = improve_5throot_solution_sse2(x, arg4);
/* error max = 0.00021 avg = 1.6e-05 |avg| = 1.6e-05 */
x = improve_5throot_solution(x, arg4);
x = improve_5throot_solution_sse2(x, arg4);
/* error max = 6.1e-07 avg = 5.2e-08 |avg| = 1.1e-07 */
x = improve_5throot_solution(x, arg4);
x = improve_5throot_solution_sse2(x, arg4);
return x * (x * x);
}
ccl_device float4 color_srgb_to_linear(const float4 &c)
ccl_device float4 color_srgb_to_linear_sse2(const float4 &c)
{
int4 cmp = c < make_float4(0.04045f);
float4 lt = max(c * make_float4(1.0f / 12.92f), make_float4(0.0f));
float4 gtebase = (c + make_float4(0.055f)) * make_float4(1.0f / 1.055f); /* fma */
float4 gte = fastpow24(gtebase);
float4 gte = fastpow24_sse2(gtebase);
return select(cmp, lt, gte);
}
#endif /* __KERNEL_SSE2__ */
@@ -328,7 +328,7 @@ ccl_device float4 color_srgb_to_linear_v4(float4 c)
{
#ifdef __KERNEL_SSE2__
float4 r = c;
r = color_srgb_to_linear(r);
r = color_srgb_to_linear_sse2(r);
r.w = c.w;
return r;
#else