MSL: Improve handling of sample masks.

ppenenko · Oct 23, 2023 · 2fba284 · 2fba284
1 parent 56bdcfa
commit 2fba284
Show file tree

Hide file tree

Showing 14 changed files with 124 additions and 23 deletions.
diff --git a/reference/opt/shaders-msl/frag/post-depth-coverage.ios.msl2.frag b/reference/opt/shaders-msl/frag/post-depth-coverage.ios.msl2.frag
@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 
diff --git a/reference/opt/shaders-msl/frag/post-depth-coverage.msl23.frag b/reference/opt/shaders-msl/frag/post-depth-coverage.msl23.frag
@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 
diff --git a/reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag b/reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag
@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]], uint gl_SampleID
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22 & (1 << gl_SampleID));
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22 & (1 << gl_SampleID)));
     out.gl_SampleMask &= 0x22;
     return out;
 }

diff --git a/reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag b/reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag
@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22);
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22));
     out.gl_SampleMask &= 0x22;
     return out;
 }

diff --git a/reference/opt/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag b/reference/opt/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag
@@ -1,8 +1,49 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
 #include <metal_stdlib>
 #include <simd/simd.h>
 
 using namespace metal;
 
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
 struct type_View
 {
     float4x4 View_TranslatedWorldToClip;
@@ -489,12 +530,12 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     if (View.View_NumSceneColorMSAASamples > 1)
     {
         _268 = _255 * float4(float(View.View_NumSceneColorMSAASamples) * 0.25);
-        _269 = gl_SampleMaskIn & 15u;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0] & 15u;
     }
     else
     {
         _268 = _255;
-        _269 = gl_SampleMaskIn;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0];
     }
     out.out_var_SV_Target0 = _268;
     out.gl_SampleMask = _269;

diff --git a/reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag b/reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag
@@ -52,8 +52,8 @@ struct main0_out
 fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
-    spvUnsafeArray<uint, 1> copy_sample_mask = gl_SampleMaskIn;
-    out.gl_SampleMask = copy_sample_mask;
+    spvUnsafeArray<uint, 1> copy_sample_mask = spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) });
+    out.gl_SampleMask = copy_sample_mask[0];
     return out;
 }
 
diff --git a/reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag b/reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag
@@ -52,8 +52,8 @@ struct main0_out
 fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
-    spvUnsafeArray<int, 1> copy_sample_mask = gl_SampleMaskIn;
-    out.gl_SampleMask = copy_sample_mask;
+    spvUnsafeArray<int, 1> copy_sample_mask = spvUnsafeArray<int, 1>({ int(gl_SampleMaskIn) });
+    out.gl_SampleMask = copy_sample_mask[0];
     return out;
 }
 
diff --git a/reference/shaders-msl/frag/post-depth-coverage.ios.msl2.frag b/reference/shaders-msl/frag/post-depth-coverage.ios.msl2.frag
@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 
diff --git a/reference/shaders-msl/frag/post-depth-coverage.msl23.frag b/reference/shaders-msl/frag/post-depth-coverage.msl23.frag
@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 
diff --git a/reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag b/reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag
@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]], uint gl_SampleID
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22 & (1 << gl_SampleID));
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22 & (1 << gl_SampleID)));
     out.gl_SampleMask &= 0x22;
     return out;
 }

diff --git a/reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag b/reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag
@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22);
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22));
     out.gl_SampleMask &= 0x22;
     return out;
 }

diff --git a/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag b/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag
@@ -1,8 +1,49 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
 #include <metal_stdlib>
 #include <simd/simd.h>
 
 using namespace metal;
 
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
 struct type_View
 {
     float4x4 View_TranslatedWorldToClip;
@@ -489,12 +530,12 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     if (View.View_NumSceneColorMSAASamples > 1)
     {
         _268 = _255 * float4(float(View.View_NumSceneColorMSAASamples) * 0.25);
-        _269 = gl_SampleMaskIn & 15u;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0] & 15u;
     }
     else
     {
         _268 = _255;
-        _269 = gl_SampleMaskIn;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0];
     }
     out.out_var_SV_Target0 = _268;
     out.gl_SampleMask = _269;

diff --git a/spirv_glsl.cpp b/spirv_glsl.cpp
@@ -10158,10 +10158,11 @@ string CompilerGLSL::access_chain_internal(uint32_t base, const uint32_t *indice
 				if (!pending_array_enclose)
 					expr += "]";
 			}
-			// Some builtins are arrays in SPIR-V but not in other languages, e.g. gl_SampleMask[] is an array in SPIR-V but not in Metal.
-			// By throwing away the index, we imply the index was 0, which it must be for gl_SampleMask.
-			else if (!builtin_translates_to_nonarray(BuiltIn(get_decoration(base, DecorationBuiltIn))))
+			else if (index_is_literal || !builtin_translates_to_nonarray(BuiltIn(get_decoration(base, DecorationBuiltIn))))
 			{
+				// Some builtins are arrays in SPIR-V but not in other languages, e.g. gl_SampleMask[] is an array in SPIR-V but not in Metal.
+				// By throwing away the index, we imply the index was 0, which it must be for gl_SampleMask.
+				// For literal indices we are working on composites, so we ignore this since we have already converted to proper array.
 				append_index(index, is_literal);
 			}
 

diff --git a/spirv_msl.cpp b/spirv_msl.cpp
@@ -4911,9 +4911,18 @@ void CompilerMSL::emit_store_statement(uint32_t lhs_expression, uint32_t rhs_exp
 
 	bool transpose = lhs_e && lhs_e->need_transpose;
 
-	// No physical type remapping, and no packed type, so can just emit a store directly.
-	if (!lhs_remapped_type && !lhs_packed_type)
+	if (has_decoration(lhs_expression, DecorationBuiltIn) &&
+	    BuiltIn(get_decoration(lhs_expression, DecorationBuiltIn)) == BuiltInSampleMask &&
+	    type_is_top_level_array(type))
 	{
+		// Storing an array to SampleMask, have to remove the array-ness before storing.
+		statement(to_expression(lhs_expression), " = ", to_enclosed_unpacked_expression(rhs_expression), "[0];");
+		register_write(lhs_expression);
+	}
+	else if (!lhs_remapped_type && !lhs_packed_type)
+	{
+		// No physical type remapping, and no packed type, so can just emit a store directly.
+
 		// We might not be dealing with remapped physical types or packed types,
 		// but we might be doing a clean store to a row-major matrix.
 		// In this case, we just flip transpose states, and emit the store, a transpose must be in the RHS expression, if any.
@@ -17178,6 +17187,7 @@ void CompilerMSL::cast_from_variable_load(uint32_t source_id, std::string &expr,
 	case BuiltInInstanceIndex:
 	case BuiltInBaseInstance:
 	case BuiltInBaseVertex:
+	case BuiltInSampleMask:
 		expected_type = SPIRType::UInt;
 		expected_width = 32;
 		break;
@@ -17195,9 +17205,17 @@ void CompilerMSL::cast_from_variable_load(uint32_t source_id, std::string &expr,
 		break;
 	}
 
-	if (expected_type != expr_type.basetype)
+	if (type_is_top_level_array(expr_type) && builtin == BuiltInSampleMask)
+	{
+		// Needs special handling.
+		auto wrap_expr = join(type_to_glsl(expr_type), "({ ");
+		wrap_expr += join(type_to_glsl(get<SPIRType>(expr_type.parent_type)), "(", expr, ")");
+		wrap_expr += " })";
+		expr = std::move(wrap_expr);
+	}
+	else if (expected_type != expr_type.basetype)
 	{
-		if (!expr_type.array.empty() && (builtin == BuiltInTessLevelInner || builtin == BuiltInTessLevelOuter))
+		if (type_is_top_level_array(expr_type) && (builtin == BuiltInTessLevelInner || builtin == BuiltInTessLevelOuter))
 		{
 			// Triggers when loading TessLevel directly as an array.
 			// Need explicit padding + cast.