From 8077384e3a2fd3d90ef64f3e4e9f568fad979da8 Mon Sep 17 00:00:00 2001
From: Michael Jones <michael_p_jones@apple.com>
Date: Fri, 18 Jul 2025 11:18:43 +0200
Subject: [PATCH] Cycles: Improve Metal kernel specialisation

This improves the existing scene specialisation mechanism by replacing "kernel_data.kernel_features" with a function constant. It doesn't cause any additional compilation requests, but allows the backend compiler to eliminate more dead code. An additional compiler hint is provided for dead-stripping "volume_stack_enter_exit" which results in slightly faster rendering of non-volumetric scenes.

Pull Request: https://projects.blender.org/blender/blender/pulls/142235
---
 intern/cycles/device/metal/device_impl.mm              | 4 ++++
 intern/cycles/device/metal/kernel.mm                   | 4 ++++
 intern/cycles/kernel/device/metal/function_constants.h | 4 ++++
 intern/cycles/kernel/integrator/volume_stack.h         | 9 +++++++++
 4 files changed, 21 insertions(+)

diff --git a/intern/cycles/device/metal/device_impl.mm b/intern/cycles/device/metal/device_impl.mm
index 218183a0af4..e7f9151d952 100644
--- a/intern/cycles/device/metal/device_impl.mm
+++ b/intern/cycles/device/metal/device_impl.mm
@@ -309,6 +309,10 @@ string MetalDevice::preprocess_source(MetalPipelineType pso_type,
 #  undef KERNEL_STRUCT_MEMBER_DONT_SPECIALIZE
 #  undef KERNEL_STRUCT_BEGIN
 
+      /* Replace "kernel_data.kernel_features" memory fetches with a function constant. */
+      string_replace_same_length(
+          *source, "kernel_data.kernel_features", "kernel_data_kernel_features");
+
       metal_printf("KernelData patching took %.1f ms", (time_dt() - starttime) * 1000.0);
     }
 
diff --git a/intern/cycles/device/metal/kernel.mm b/intern/cycles/device/metal/kernel.mm
index 79e90526a27..64ddf316d6a 100644
--- a/intern/cycles/device/metal/kernel.mm
+++ b/intern/cycles/device/metal/kernel.mm
@@ -461,6 +461,10 @@ static MTLFunctionConstantValues *GetConstantValues(const KernelData *data = nul
 
 #  include "kernel/data_template.h"
 
+  [constant_values setConstantValue:&data->kernel_features
+                               type:MTLDataTypeInt
+                            atIndex:KernelData_kernel_features];
+
   return constant_values;
 }
 
diff --git a/intern/cycles/kernel/device/metal/function_constants.h b/intern/cycles/kernel/device/metal/function_constants.h
index 17fb034300d..4fed68bd199 100644
--- a/intern/cycles/kernel/device/metal/function_constants.h
+++ b/intern/cycles/kernel/device/metal/function_constants.h
@@ -6,6 +6,8 @@ enum {
   Kernel_DummyConstant,
 #define KERNEL_STRUCT_MEMBER(parent, type, name) KernelData_##parent##_##name,
 #include "kernel/data_template.h"
+
+  KernelData_kernel_features
 };
 
 #ifdef __KERNEL_METAL__
@@ -13,4 +15,6 @@ enum {
     constant type kernel_data_##parent##_##name \
         [[function_constant(KernelData_##parent##_##name)]];
 #  include "kernel/data_template.h"
+
+constant int kernel_data_kernel_features [[function_constant(KernelData_kernel_features)]];
 #endif
diff --git a/intern/cycles/kernel/integrator/volume_stack.h b/intern/cycles/kernel/integrator/volume_stack.h
index 95f84be385f..094a38caaea 100644
--- a/intern/cycles/kernel/integrator/volume_stack.h
+++ b/intern/cycles/kernel/integrator/volume_stack.h
@@ -27,6 +27,15 @@ ccl_device void volume_stack_enter_exit(KernelGlobals kg,
                                         StackReadOp stack_read,
                                         StackWriteOp stack_write)
 {
+#  ifdef __KERNEL_USE_DATA_CONSTANTS__
+  /* If we're using data constants, this fetch disappears.
+   * On Apple GPUs, scenes without volumetric features can render 1 or 2% faster by dead-stripping
+   * this function. */
+  if (!(kernel_data.kernel_features & KERNEL_FEATURE_VOLUME)) {
+    return;
+  }
+#  endif
+
   /* todo: we should have some way for objects to indicate if they want the
    * world shader to work inside them. excluding it by default is problematic
    * because non-volume objects can't be assumed to be closed manifolds */