[Bf-blender-cvs] [f9f834068e3] blender-v3.3-release: Cycles: Allow Intel GPUs under Metal

[Morteza Mostajab]

Commit: f9f834068e38f562f3590c12af35d3033db98995
Author: Morteza Mostajab
Date:   Wed Oct 19 17:09:23 2022 +0100
Branches: blender-v3.3-release
https://developer.blender.org/rBf9f834068e38f562f3590c12af35d3033db98995

Cycles: Allow Intel GPUs under Metal

Known Issues:
- Command buffer failures when using binary archives (binary archives is disabled for Intel GPUs as a workaround)
- Wrong texture sampler being applied (to be addressed in the future)

Ref T92212

Differential Revision: https://developer.blender.org/D16253

===================================================================

M	intern/cycles/device/metal/device_impl.mm
M	intern/cycles/device/metal/kernel.mm
M	intern/cycles/device/metal/util.mm
M	intern/cycles/kernel/device/metal/context_begin.h

===================================================================

diff --git a/intern/cycles/device/metal/device_impl.mm b/intern/cycles/device/metal/device_impl.mm
index 6feeaa0707c..f40b056606f 100644
--- a/intern/cycles/device/metal/device_impl.mm
+++ b/intern/cycles/device/metal/device_impl.mm
@@ -338,6 +338,12 @@ bool MetalDevice::compile_and_load(MetalPipelineType pso_type)
 
   MTLCompileOptions *options = [[MTLCompileOptions alloc] init];
 
+  if (@available(macos 13.0, *)) {
+    if (device_vendor == METAL_GPU_INTEL) {
+      [options setOptimizationLevel:MTLLibraryOptimizationLevelSize];
+    }
+  }
+
   options.fastMathEnabled = YES;
   if (@available(macOS 12.0, *)) {
     options.languageVersion = MTLLanguageVersion2_4;
diff --git a/intern/cycles/device/metal/kernel.mm b/intern/cycles/device/metal/kernel.mm
index e22b0159108..e06858db3ce 100644
--- a/intern/cycles/device/metal/kernel.mm
+++ b/intern/cycles/device/metal/kernel.mm
@@ -325,6 +325,12 @@ bool MetalKernelPipeline::should_use_binary_archive() const
       }
     }
 
+    /* Workaround for Intel GPU having issue using Binary Archives */
+    MetalGPUVendor gpu_vendor = MetalInfo::get_device_vendor(mtlDevice);
+    if (gpu_vendor == METAL_GPU_INTEL) {
+      return false;
+    }
+
     if (pso_type == PSO_GENERIC) {
       /* Archive the generic kernels. */
       return true;
diff --git a/intern/cycles/device/metal/util.mm b/intern/cycles/device/metal/util.mm
index 65c67c400fe..eb77aeb6a54 100644
--- a/intern/cycles/device/metal/util.mm
+++ b/intern/cycles/device/metal/util.mm
@@ -110,6 +110,10 @@ vector<id<MTLDevice>> const &MetalInfo::get_usable_devices()
       usable |= (vendor == METAL_GPU_AMD);
     }
 
+    if (@available(macos 13.0, *)) {
+      usable |= (vendor == METAL_GPU_INTEL);
+    }
+
     if (usable) {
       metal_printf("- %s\n", device_name.c_str());
       [device retain];
diff --git a/intern/cycles/kernel/device/metal/context_begin.h b/intern/cycles/kernel/device/metal/context_begin.h
index 99cb1e3826e..e75ec9cadec 100644
--- a/intern/cycles/kernel/device/metal/context_begin.h
+++ b/intern/cycles/kernel/device/metal/context_begin.h
@@ -34,21 +34,48 @@ class MetalKernelContext {
       kernel_assert(0);
       return 0;
     }
-
+    
+#ifdef __KERNEL_METAL_INTEL__
+    template<typename TextureType, typename CoordsType>
+    inline __attribute__((__always_inline__))
+    auto ccl_gpu_tex_object_read_intel_workaround(TextureType texture_array,
+                                                  const uint tid, const uint sid,
+                                                  CoordsType coords) const
+    {
+      switch(sid) {
+        default:
+        case 0: return texture_array[tid].tex.sample(sampler(address::repeat, filter::nearest), coords);
+        case 1: return texture_array[tid].tex.sample(sampler(address::clamp_to_edge, filter::nearest), coords);
+        case 2: return texture_array[tid].tex.sample(sampler(address::clamp_to_zero, filter::nearest), coords);
+        case 3: return texture_array[tid].tex.sample(sampler(address::repeat, filter::linear), coords);
+        case 4: return texture_array[tid].tex.sample(sampler(address::clamp_to_edge, filter::linear), coords);
+        case 5: return texture_array[tid].tex.sample(sampler(address::clamp_to_zero, filter::linear), coords);
+      }
+    }
+#endif
+    
     // texture2d
     template<>
     inline __attribute__((__always_inline__))
     float4 ccl_gpu_tex_object_read_2D(ccl_gpu_tex_object_2D tex, float x, float y) const {
       const uint tid(tex);
       const uint sid(tex >> 32);
+#ifndef __KERNEL_METAL_INTEL__
       return metal_ancillaries->textures_2d[tid].tex.sample(metal_samplers[sid], float2(x, y));
+#else
+      return ccl_gpu_tex_object_read_intel_workaround(metal_ancillaries->textures_2d, tid, sid, float2(x, y));
+#endif
     }
     template<>
     inline __attribute__((__always_inline__))
     float ccl_gpu_tex_object_read_2D(ccl_gpu_tex_object_2D tex, float x, float y) const {
       const uint tid(tex);
       const uint sid(tex >> 32);
+#ifndef __KERNEL_METAL_INTEL__
       return metal_ancillaries->textures_2d[tid].tex.sample(metal_samplers[sid], float2(x, y)).x;
+#else
+      return ccl_gpu_tex_object_read_intel_workaround(metal_ancillaries->textures_2d, tid, sid, float2(x, y)).x;
+#endif
     }
 
     // texture3d
@@ -57,14 +84,22 @@ class MetalKernelContext {
     float4 ccl_gpu_tex_object_read_3D(ccl_gpu_tex_object_3D tex, float x, float y, float z) const {
       const uint tid(tex);
       const uint sid(tex >> 32);
+#ifndef __KERNEL_METAL_INTEL__
       return metal_ancillaries->textures_3d[tid].tex.sample(metal_samplers[sid], float3(x, y, z));
+#else
+      return ccl_gpu_tex_object_read_intel_workaround(metal_ancillaries->textures_3d, tid, sid, float3(x, y, z));
+#endif
     }
     template<>
     inline __attribute__((__always_inline__))
     float ccl_gpu_tex_object_read_3D(ccl_gpu_tex_object_3D tex, float x, float y, float z) const {
       const uint tid(tex);
       const uint sid(tex >> 32);
+#ifndef __KERNEL_METAL_INTEL__
       return metal_ancillaries->textures_3d[tid].tex.sample(metal_samplers[sid], float3(x, y, z)).x;
+#else
+      return ccl_gpu_tex_object_read_intel_workaround(metal_ancillaries->textures_3d, tid, sid, float3(x, y, z)).x;
+#endif
     }
 #    include "kernel/device/gpu/image.h"