Cycles: Adapt volumetric lambda functions to work on MSL

This patch adapts the existing volumetric read/write lambda functions for Metal. Lambda expressions are not supported on MSL, so two new macros `VOLUME_READ_LAMBDA` and `VOLUME_WRITE_LAMBDA` have been defined with a default implementation which, on Metal, is overridden to use inline function objects.

This patch also removes the last remaining mention of the now-unused `ccl_addr_space`.

Ref T92212

Reviewed By: leesonw

Maniphest Tasks: T92212

Differential Revision: https://developer.blender.org/D13234
This commit is contained in:
Michael Jones 2021-11-16 13:41:29 +00:00 committed by Michael Jones
parent 85ac9b8584
commit 64003fa4b0
5 changed files with 47 additions and 31 deletions

@ -150,6 +150,31 @@ void kernel_gpu_##name::run(thread MetalKernelContext& context, \
// clang-format on
/* volumetric lambda functions - use function objects for lambda-like functionality */
#define VOLUME_READ_LAMBDA(function_call) \
struct FnObjectRead { \
KernelGlobals kg; \
ccl_private MetalKernelContext *context; \
int state; \
\
VolumeStack operator()(const int i) const \
{ \
return context->function_call; \
} \
} volume_read_lambda_pass{kg, this, state};
#define VOLUME_WRITE_LAMBDA(function_call) \
struct FnObjectWrite { \
KernelGlobals kg; \
ccl_private MetalKernelContext *context; \
int state; \
\
void operator()(const int i, VolumeStack entry) const \
{ \
context->function_call; \
} \
} volume_write_lambda_pass{kg, this, state};
/* make_type definitions with Metal style element initializers */
#ifdef make_float2
# undef make_float2

@ -95,8 +95,8 @@ ccl_device_inline void integrate_transparent_volume_shadow(KernelGlobals kg,
shader_setup_from_volume(kg, shadow_sd, &ray);
const float step_size = volume_stack_step_size(
kg, [=](const int i) { return integrator_state_read_shadow_volume_stack(state, i); });
VOLUME_READ_LAMBDA(integrator_state_read_shadow_volume_stack(state, i));
const float step_size = volume_stack_step_size(kg, volume_read_lambda_pass);
volume_shadow_heterogeneous(kg, state, &ray, shadow_sd, throughput, step_size);
}

@ -78,9 +78,8 @@ ccl_device_inline bool shadow_volume_shader_sample(KernelGlobals kg,
ccl_private ShaderData *ccl_restrict sd,
ccl_private float3 *ccl_restrict extinction)
{
shader_eval_volume<true>(kg, state, sd, PATH_RAY_SHADOW, [=](const int i) {
return integrator_state_read_shadow_volume_stack(state, i);
});
VOLUME_READ_LAMBDA(integrator_state_read_shadow_volume_stack(state, i))
shader_eval_volume<true>(kg, state, sd, PATH_RAY_SHADOW, volume_read_lambda_pass);
if (!(sd->flag & SD_EXTINCTION)) {
return false;
@ -98,9 +97,8 @@ ccl_device_inline bool volume_shader_sample(KernelGlobals kg,
ccl_private VolumeShaderCoefficients *coeff)
{
const uint32_t path_flag = INTEGRATOR_STATE(state, path, flag);
shader_eval_volume<false>(kg, state, sd, path_flag, [=](const int i) {
return integrator_state_read_volume_stack(state, i);
});
VOLUME_READ_LAMBDA(integrator_state_read_volume_stack(state, i))
shader_eval_volume<false>(kg, state, sd, path_flag, volume_read_lambda_pass);
if (!(sd->flag & (SD_EXTINCTION | SD_SCATTER | SD_EMISSION))) {
return false;
@ -921,8 +919,8 @@ ccl_device VolumeIntegrateEvent volume_integrate(KernelGlobals kg,
VOLUME_SAMPLE_DISTANCE;
/* Step through volume. */
const float step_size = volume_stack_step_size(
kg, [=](const int i) { return integrator_state_read_volume_stack(state, i); });
VOLUME_READ_LAMBDA(integrator_state_read_volume_stack(state, i))
const float step_size = volume_stack_step_size(kg, volume_read_lambda_pass);
/* TODO: expensive to zero closures? */
VolumeIntegrateResult result = {};

@ -18,6 +18,14 @@
CCL_NAMESPACE_BEGIN
/* Volumetric read/write lambda functions - default implementations */
#ifndef VOLUME_READ_LAMBDA
# define VOLUME_READ_LAMBDA(function_call) \
auto volume_read_lambda_pass = [=](const int i) { return function_call; };
# define VOLUME_WRITE_LAMBDA(function_call) \
auto volume_write_lambda_pass = [=](const int i, VolumeStack entry) { function_call; };
#endif
/* Volume Stack
*
* This is an array of object/shared ID's that the current segment of the path
@ -88,26 +96,18 @@ ccl_device void volume_stack_enter_exit(KernelGlobals kg,
IntegratorState state,
ccl_private const ShaderData *sd)
{
volume_stack_enter_exit(
kg,
sd,
[=](const int i) { return integrator_state_read_volume_stack(state, i); },
[=](const int i, const VolumeStack entry) {
integrator_state_write_volume_stack(state, i, entry);
});
VOLUME_READ_LAMBDA(integrator_state_read_volume_stack(state, i))
VOLUME_WRITE_LAMBDA(integrator_state_write_volume_stack(state, i, entry))
volume_stack_enter_exit(kg, sd, volume_read_lambda_pass, volume_write_lambda_pass);
}
ccl_device void shadow_volume_stack_enter_exit(KernelGlobals kg,
IntegratorShadowState state,
ccl_private const ShaderData *sd)
{
volume_stack_enter_exit(
kg,
sd,
[=](const int i) { return integrator_state_read_shadow_volume_stack(state, i); },
[=](const int i, const VolumeStack entry) {
integrator_state_write_shadow_volume_stack(state, i, entry);
});
VOLUME_READ_LAMBDA(integrator_state_read_shadow_volume_stack(state, i))
VOLUME_WRITE_LAMBDA(integrator_state_write_shadow_volume_stack(state, i, entry))
volume_stack_enter_exit(kg, sd, volume_read_lambda_pass, volume_write_lambda_pass);
}
/* Clean stack after the last bounce.

@ -36,13 +36,6 @@
# define __KERNEL_CPU__
#endif
/* TODO(sergey): This is only to make it possible to include this header
* from outside of the kernel. but this could be done somewhat cleaner?
*/
#ifndef ccl_addr_space
# define ccl_addr_space
#endif
CCL_NAMESPACE_BEGIN
/* Constants */