Cleanup: remove __KERNEL_CPU__

This was tested in some places to check if code was being compiled for the
CPU, however this is only defined in the kernel. Checking __KERNEL_GPU__
always works.

intern/cycles/device/optix/device_impl.cpp

@@ -26,7 +26,6 @@
 #  include "util/task.h"
 #  include "util/time.h"
 
-#  undef __KERNEL_CPU__
 #  define __KERNEL_OPTIX__
 #  include "kernel/device/optix/globals.h"
 

intern/cycles/device/optix/queue.cpp

@@ -8,7 +8,6 @@
 
 #  include "util/time.h"
 
-#  undef __KERNEL_CPU__
 #  define __KERNEL_OPTIX__
 #  include "kernel/device/optix/globals.h"
 

intern/cycles/kernel/bvh/util.h

@@ -33,7 +33,7 @@ ccl_device_forceinline float intersection_t_offset(const float t)
   return __uint_as_float(bits);
 }
 
-#if defined(__KERNEL_CPU__)
+#ifndef __KERNEL_GPU__
 ccl_device int intersections_compare(const void *a, const void *b)
 {
   const Intersection *isect_a = (const Intersection *)a;

intern/cycles/kernel/closure/bsdf_hair_principled.h

@@ -3,7 +3,7 @@
 
 #pragma once
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 #  include <fenv.h>
 #endif
 

intern/cycles/kernel/device/cpu/compat.h

@@ -3,8 +3,6 @@
 
 #pragma once
 
-#define __KERNEL_CPU__
-
 /* Release kernel has too much false-positive maybe-uninitialized warnings,
  * which makes it possible to miss actual warnings.
  */

intern/cycles/kernel/integrator/intersect_shadow.h

@@ -51,7 +51,7 @@ ccl_device_forceinline int integrate_shadow_max_transparent_hits(KernelGlobals k
 }
 
 #ifdef __TRANSPARENT_SHADOWS__
-#  if defined(__KERNEL_CPU__)
+#  ifndef __KERNEL_GPU__
 ccl_device int shadow_intersections_compare(const void *a, const void *b)
 {
   const Intersection *isect_a = (const Intersection *)a;

intern/cycles/kernel/integrator/path_state.h

@@ -13,7 +13,7 @@ CCL_NAMESPACE_BEGIN
 ccl_device_inline void path_state_init_queues(IntegratorState state)
 {
   INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = 0;
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
   INTEGRATOR_STATE_WRITE(&state->shadow, shadow_path, queued_kernel) = 0;
   INTEGRATOR_STATE_WRITE(&state->ao, shadow_path, queued_kernel) = 0;
 #endif

intern/cycles/kernel/integrator/state.h

@@ -140,7 +140,7 @@ typedef struct IntegratorStateGPU {
  * happen from a kernel which operates on a "main" path. Attempt to use shadow catcher accessors
  * from a kernel which operates on a shadow catcher state will cause bad memory access. */
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 
 /* Scalar access on CPU. */
 
@@ -159,7 +159,7 @@ typedef const IntegratorShadowStateCPU *ccl_restrict ConstIntegratorShadowState;
 #  define INTEGRATOR_STATE_ARRAY_WRITE(state, nested_struct, array_index, member) \
     ((state)->nested_struct[array_index].member)
 
-#else /* __KERNEL_CPU__ */
+#else /* !__KERNEL_GPU__ */
 
 /* Array access on GPU with Structure-of-Arrays. */
 
@@ -180,6 +180,6 @@ typedef int ConstIntegratorShadowState;
 #  define INTEGRATOR_STATE_ARRAY_WRITE(state, nested_struct, array_index, member) \
     INTEGRATOR_STATE_ARRAY(state, nested_struct, array_index, member)
 
-#endif /* __KERNEL_CPU__ */
+#endif /* !__KERNEL_GPU__ */
 
 CCL_NAMESPACE_END

intern/cycles/kernel/integrator/state_util.h

@@ -338,7 +338,7 @@ ccl_device_inline IntegratorState integrator_state_shadow_catcher_split(KernelGl
   return to_state;
 }
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 ccl_device_inline int integrator_state_bounce(ConstIntegratorState state, const int)
 {
   return INTEGRATOR_STATE(state, path, bounce);

intern/cycles/kernel/types.h

@@ -19,10 +19,6 @@
 
 #include "kernel/svm/types.h"
 
-#ifndef __KERNEL_GPU__
-#  define __KERNEL_CPU__
-#endif
-
 CCL_NAMESPACE_BEGIN
 
 /* Constants */
@@ -51,10 +47,10 @@ CCL_NAMESPACE_BEGIN
 #define INTEGRATOR_SHADOW_ISECT_SIZE_CPU 1024U
 #define INTEGRATOR_SHADOW_ISECT_SIZE_GPU 4U
 
-#ifdef __KERNEL_CPU__
-#  define INTEGRATOR_SHADOW_ISECT_SIZE INTEGRATOR_SHADOW_ISECT_SIZE_CPU
-#else
+#ifdef __KERNEL_GPU__
 #  define INTEGRATOR_SHADOW_ISECT_SIZE INTEGRATOR_SHADOW_ISECT_SIZE_GPU
+#else
+#  define INTEGRATOR_SHADOW_ISECT_SIZE INTEGRATOR_SHADOW_ISECT_SIZE_CPU
 #endif
 
 /* Kernel features */
@@ -91,12 +87,12 @@ CCL_NAMESPACE_BEGIN
 #define __BRANCHED_PATH__
 
 /* Device specific features */
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 #  ifdef WITH_OSL
 #    define __OSL__
 #  endif
 #  define __VOLUME_RECORD_ALL__
-#endif /* __KERNEL_CPU__ */
+#endif /* !__KERNEL_GPU__ */
 
 /* MNEE currently causes "Compute function exceeds available temporary registers"
  * on Metal, disabled for now. */
@@ -722,7 +718,7 @@ typedef struct ccl_align(16) ShaderClosure
 {
   SHADER_CLOSURE_BASE;
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
   float pad[2];
 #endif
   float data[10];
@@ -1540,15 +1536,15 @@ enum KernelFeatureFlag : uint32_t {
 /* Must be constexpr on the CPU to avoid compile errors because the state types
  * are different depending on the main, shadow or null path. For GPU we don't have
  * C++17 everywhere so can't use it. */
-#ifdef __KERNEL_CPU__
+#ifdef __KERNEL_GPU__
+#  define IF_KERNEL_FEATURE(feature) if ((node_feature_mask & (KERNEL_FEATURE_##feature)) != 0U)
+#  define IF_KERNEL_NODES_FEATURE(feature) \
+    if ((node_feature_mask & (KERNEL_FEATURE_NODE_##feature)) != 0U)
+#else
 #  define IF_KERNEL_FEATURE(feature) \
     if constexpr ((node_feature_mask & (KERNEL_FEATURE_##feature)) != 0U)
 #  define IF_KERNEL_NODES_FEATURE(feature) \
     if constexpr ((node_feature_mask & (KERNEL_FEATURE_NODE_##feature)) != 0U)
-#else
-#  define IF_KERNEL_FEATURE(feature) if ((node_feature_mask & (KERNEL_FEATURE_##feature)) != 0U)
-#  define IF_KERNEL_NODES_FEATURE(feature) \
-    if ((node_feature_mask & (KERNEL_FEATURE_NODE_##feature)) != 0U)
 #endif
 
 CCL_NAMESPACE_END

intern/cycles/kernel/util/profiling.h

@@ -3,13 +3,13 @@
 
 #pragma once
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 #  include "util/profiling.h"
 #endif
 
 CCL_NAMESPACE_BEGIN
 
-#ifdef __KERNEL_CPU__
+#ifndef __KERNEL_GPU__
 #  define PROFILING_INIT(kg, event) \
     ProfilingHelper profiling_helper((ProfilingState *)&kg->profiler, event)
 #  define PROFILING_EVENT(event) profiling_helper.set_event(event)
@@ -22,6 +22,6 @@ CCL_NAMESPACE_BEGIN
 #  define PROFILING_EVENT(event)
 #  define PROFILING_INIT_FOR_SHADER(kg, event)
 #  define PROFILING_SHADER(object, shader)
-#endif /* __KERNEL_CPU__ */
+#endif /* !__KERNEL_GPU__ */
 
 CCL_NAMESPACE_END

intern/cycles/test/util_avxf_avx2_test.cpp

@@ -2,7 +2,6 @@
  * Copyright 2011-2022 Blender Foundation */
 
 #define __KERNEL_AVX2__
-#define __KERNEL_CPU__
 
 #define TEST_CATEGORY_NAME util_avx2
 

intern/cycles/test/util_avxf_avx_test.cpp

@@ -2,7 +2,6 @@
  * Copyright 2011-2022 Blender Foundation */
 
 #define __KERNEL_AVX__
-#define __KERNEL_CPU__
 
 #define TEST_CATEGORY_NAME util_avx
 

intern/cycles/util/defines.h

@@ -81,7 +81,7 @@
 /* macros */
 
 /* hints for branch prediction, only use in code that runs a _lot_ */
-#if defined(__GNUC__) && defined(__KERNEL_CPU__)
+#if defined(__GNUC__) && !defined(__KERNEL_GPU__)
 #  define LIKELY(x) __builtin_expect(!!(x), 1)
 #  define UNLIKELY(x) __builtin_expect(!!(x), 0)
 #else

intern/cycles/util/math_fast.h

@@ -420,7 +420,7 @@ ccl_device_inline float fast_expf(float x)
   return fast_exp2f(x / M_LN2_F);
 }
 
-#if defined(__KERNEL_CPU__) && !defined(_MSC_VER)
+#if !defined(__KERNEL_GPU__) && !defined(_MSC_VER)
 /* MSVC seems to have a code-gen bug here in at least SSE41/AVX, see
  * T78047 and T78869 for details. Just disable for now, it only makes
  * a small difference in denoising performance. */