MSL: Prevent RAW hazards on read_write textures

5 files changed, 9 insertions, 1 deletions

diff --git a/reference/opt/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp b/reference/opt/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
index 53655639..346f40ee 100644
--- a/reference/opt/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
+++ b/reference/opt/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
@@ -8,6 +8,7 @@ using namespace metal;
 static inline __attribute__((always_inline))
 void _main(thread const uint3& id, texture2d<float, access::read_write> TargetTexture)
 {
+    TargetTexture.fence();
     float2 loaded = TargetTexture.read(uint2(id.xy)).xy;
     float2 storeTemp = loaded + float2(1.0);
     TargetTexture.write(storeTemp.xyyy, uint2((id.xy + uint2(1u))));
diff --git a/reference/opt/shaders-msl/desktop-only/frag/image-ms.desktop.frag b/reference/opt/shaders-msl/desktop-only/frag/image-ms.desktop.frag
index 4083e4ea..30046a1f 100644
--- a/reference/opt/shaders-msl/desktop-only/frag/image-ms.desktop.frag
+++ b/reference/opt/shaders-msl/desktop-only/frag/image-ms.desktop.frag
@@ -5,6 +5,7 @@ using namespace metal;
 
 fragment void main0(texture2d_ms<float> uImageMS [[texture(0)]], texture2d_array<float, access::read_write> uImageArray [[texture(1)]], texture2d<float, access::write> uImage [[texture(2)]])
 {
+    uImageArray.fence();
     uImage.write(uImageMS.read(uint2(int2(1, 2)), 2), uint2(int2(2, 3)));
     uImageArray.write(uImageArray.read(uint2(int3(1, 2, 4).xy), uint(int3(1, 2, 4).z)), uint2(int3(2, 3, 7).xy), uint(int3(2, 3, 7).z));
 }
diff --git a/reference/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp b/reference/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
index 53655639..346f40ee 100644
--- a/reference/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
+++ b/reference/shaders-msl/asm/comp/image-load-store-short-vector.invalid.asm.comp
@@ -8,6 +8,7 @@ using namespace metal;
 static inline __attribute__((always_inline))
 void _main(thread const uint3& id, texture2d<float, access::read_write> TargetTexture)
 {
+    TargetTexture.fence();
     float2 loaded = TargetTexture.read(uint2(id.xy)).xy;
     float2 storeTemp = loaded + float2(1.0);
     TargetTexture.write(storeTemp.xyyy, uint2((id.xy + uint2(1u))));
diff --git a/reference/shaders-msl/desktop-only/frag/image-ms.desktop.frag b/reference/shaders-msl/desktop-only/frag/image-ms.desktop.frag
index 7957b209..d413563a 100644
--- a/reference/shaders-msl/desktop-only/frag/image-ms.desktop.frag
+++ b/reference/shaders-msl/desktop-only/frag/image-ms.desktop.frag
@@ -6,6 +6,7 @@ using namespace metal;
 fragment void main0(texture2d_ms<float> uImageMS [[texture(0)]], texture2d_array<float, access::read_write> uImageArray [[texture(1)]], texture2d<float, access::write> uImage [[texture(2)]])
 {
     float4 a = uImageMS.read(uint2(int2(1, 2)), 2);
+    uImageArray.fence();
     float4 b = uImageArray.read(uint2(int3(1, 2, 4).xy), uint(int3(1, 2, 4).z));
     uImage.write(a, uint2(int2(2, 3)));
     uImageArray.write(b, uint2(int3(2, 3, 7).xy), uint(int3(2, 3, 7).z));
diff --git a/spirv_msl.cpp b/spirv_msl.cpp
index 4ca4a04d..a8ed9b25 100644
--- a/spirv_msl.cpp
+++ b/spirv_msl.cpp
@@ -8675,9 +8675,9 @@ void CompilerMSL::emit_instruction(const Instruction &instruction)
 		// Mark that this shader reads from this image
 		uint32_t img_id = ops[2];
 		auto &type = expression_type(img_id);
+		auto *p_var = maybe_get_backing_variable(img_id);
 		if (type.image.dim != DimSubpassData)
 		{
-			auto *p_var = maybe_get_backing_variable(img_id);
 			if (p_var && has_decoration(p_var->self, DecorationNonReadable))
 			{
 				unset_decoration(p_var->self, DecorationNonReadable);
@@ -8685,6 +8685,10 @@ void CompilerMSL::emit_instruction(const Instruction &instruction)
 			}
 		}
 
+		// Metal requires explicit fences to break up RAW hazards, even within the same shader invocation
+		if (p_var && !has_decoration(p_var->self, DecorationNonWritable))
+			statement(to_expression(img_id), ".fence();");
+
 		emit_texture_op(instruction, false);
 		break;
 	}