From 7fb99a83db524a8e7c3469fd3276842f6bedfd91 Mon Sep 17 00:00:00 2001
From: Anders Leino <aleino@nvidia.com>
Date: Wed, 18 Sep 2024 15:02:41 +0300
Subject: [PATCH] Add WGSL intrinsics for synchronization

This closes issue #5085.
---
 source/slang/hlsl.meta.slang           | 18 ++++++++++++------
 source/slang/slang-capabilities.capdef |  4 ++++
 2 files changed, 16 insertions(+), 6 deletions(-)

diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang
index 7a7fd8ff06..64bb6376d3 100644
--- a/source/slang/hlsl.meta.slang
+++ b/source/slang/hlsl.meta.slang
@@ -5775,7 +5775,7 @@ bool all(matrix<T,N,M> x)
 
 // Barrier for writes to all memory spaces (HLSL SM 5.0)
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void AllMemoryBarrier()
 {
     __target_switch
@@ -5788,12 +5788,13 @@ void AllMemoryBarrier()
         {
             OpMemoryBarrier Device AcquireRelease|UniformMemory|WorkgroupMemory|ImageMemory;
         };
+    case wgsl: __intrinsic_asm "storageBarrier(); textureBarrier(); workgroupBarrier();";
     }
 }
 
 // Thread-group sync and barrier for writes to all memory spaces (HLSL SM 5.0)
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void AllMemoryBarrierWithGroupSync()
 {
     __target_switch
@@ -5806,6 +5807,7 @@ void AllMemoryBarrierWithGroupSync()
         {
             OpControlBarrier Workgroup Device AcquireRelease|UniformMemory|WorkgroupMemory|ImageMemory;
         };
+    case wgsl: __intrinsic_asm "storageBarrier(); textureBarrier(); workgroupBarrier();";
     }
 }
 
@@ -7540,7 +7542,7 @@ T determinant(matrix<T,N,N> m)
 
 // Barrier for device memory
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void DeviceMemoryBarrier()
 {
     __target_switch
@@ -7553,11 +7555,12 @@ void DeviceMemoryBarrier()
         {
             OpMemoryBarrier Device AcquireRelease|UniformMemory|ImageMemory;
         };
+    case wgsl: __intrinsic_asm "storageBarrier(); textureBarrier(); workgroupBarrier();";
     }
 }
 
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void DeviceMemoryBarrierWithGroupSync()
 {
     __target_switch
@@ -7570,6 +7573,7 @@ void DeviceMemoryBarrierWithGroupSync()
         {
             OpControlBarrier Workgroup Device AcquireRelease|UniformMemory|ImageMemory;
         };
+    case wgsl: __intrinsic_asm "storageBarrier(); textureBarrier(); workgroupBarrier();";
     }
 }
 
@@ -8932,7 +8936,7 @@ float2 GetRenderTargetSamplePosition(int Index)
 
 // Group memory barrier
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void GroupMemoryBarrier()
 {
     __target_switch
@@ -8946,6 +8950,7 @@ void GroupMemoryBarrier()
         {
             OpMemoryBarrier Workgroup AcquireRelease|WorkgroupMemory
         };
+    case wgsl: __intrinsic_asm "workgroupBarrier";
     }
 }
 
@@ -8967,7 +8972,7 @@ void __subgroupBarrier()
 }
 
 __glsl_extension(GL_KHR_memory_scope_semantics)
-[require(cuda_glsl_hlsl_metal_spirv, memorybarrier)]
+[require(cuda_glsl_hlsl_metal_spirv_wgsl, memorybarrier)]
 void GroupMemoryBarrierWithGroupSync()
 {
     __target_switch
@@ -8981,6 +8986,7 @@ void GroupMemoryBarrierWithGroupSync()
         {
             OpControlBarrier Workgroup Workgroup AcquireRelease|WorkgroupMemory
         };
+    case wgsl: __intrinsic_asm "workgroupBarrier";
     }
 }
 
diff --git a/source/slang/slang-capabilities.capdef b/source/slang/slang-capabilities.capdef
index 96f5996a0c..42a7db2133 100644
--- a/source/slang/slang-capabilities.capdef
+++ b/source/slang/slang-capabilities.capdef
@@ -322,6 +322,10 @@ alias cuda_glsl_hlsl_spirv = cuda | glsl | hlsl | spirv;
 /// [Compound]
 alias cuda_glsl_hlsl_metal_spirv = cuda | glsl | hlsl | metal | spirv;
 
+/// CUDA, GLSL, HLSL, Metal, SPIRV and WGSL code-gen targets
+/// [Compound]
+alias cuda_glsl_hlsl_metal_spirv_wgsl = cuda | glsl | hlsl | metal | spirv | wgsl;
+
 /// CUDA, GLSL, and SPIRV code-gen targets
 /// [Compound]
 alias cuda_glsl_spirv = cuda | glsl | spirv;