From 7c7d23691fc8583967a17dc501b9fb94f101990b Mon Sep 17 00:00:00 2001 From: Sergey Sharybin Date: Wed, 26 Oct 2016 14:27:31 +0200 Subject: [PATCH] Cycles: Fix crashes after recent optimization commits There is some precision issues for big magnitude coordinates which started to give weird behavior of release builds. Some weird memory usage in BVH which is tricky to nail down because only happens in release builds and GDB reports all variables as optimized out when trying to use RelWithDebInfo. There are two things in this commit: - Attempt to make vectorized code closer to original one, hoping that it'll eliminate precision issue. This seems to work for transform_point(). - Similar trick did not work for transform_direction() even tho absolute error here is much smaller. For now disabled that function, need a more careful look here. --- intern/cycles/util/util_transform.h | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) diff --git a/intern/cycles/util/util_transform.h b/intern/cycles/util/util_transform.h index 771a9442cb3..1bac3c150cb 100644 --- a/intern/cycles/util/util_transform.h +++ b/intern/cycles/util/util_transform.h @@ -84,9 +84,10 @@ ccl_device_inline float3 transform_point(const Transform *t, const float3 a) _MM_TRANSPOSE4_PS(x, y, z, w); - ssef tmp = madd(x, shuffle<0>(aa), w); - tmp = madd(y, shuffle<1>(aa), tmp); - tmp = madd(z, shuffle<2>(aa), tmp); + ssef tmp = shuffle<0>(aa) * x; + tmp = madd(shuffle<1>(aa), y, tmp); + tmp = madd(shuffle<2>(aa), z, tmp); + tmp += w; return float3(tmp.m128); #else @@ -101,7 +102,8 @@ ccl_device_inline float3 transform_point(const Transform *t, const float3 a) ccl_device_inline float3 transform_direction(const Transform *t, const float3 a) { -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE2__) + /* TODO(sergey): Disabled for now, causes crashes in certain cases. */ +#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE2__) && 0 ssef x, y, z, w, aa; aa = a.m128; x = _mm_loadu_ps(&t->x.x); @@ -111,9 +113,9 @@ ccl_device_inline float3 transform_direction(const Transform *t, const float3 a) _MM_TRANSPOSE4_PS(x, y, z, w); - ssef tmp = x * shuffle<0>(aa); - tmp = madd(y, shuffle<1>(aa), tmp); - tmp = madd(z, shuffle<2>(aa), tmp); + ssef tmp = shuffle<0>(aa) * x; + tmp = madd(shuffle<1>(aa), y, tmp); + tmp = madd(shuffle<2>(aa), z, tmp); return float3(tmp.m128); #else