MSL: Handle loading Clip/CullDistance in TESE.

Need to allow the flattened space to go through in some edge cases where
we cannot reasonably unflatten.

reference/shaders-msl-no-opt/tese/load-clip-cull.msl2.tese

@@ -0,0 +1,37 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 gl_Position [[position]];
+};
+
+struct main0_in
+{
+    float4 gl_Position [[attribute(0)]];
+    float gl_ClipDistance_0 [[attribute(1)]];
+    float gl_ClipDistance_1 [[attribute(2)]];
+    float gl_CullDistance_0 [[attribute(3)]];
+    float gl_CullDistance_1 [[attribute(4)]];
+    float gl_CullDistance_2 [[attribute(5)]];
+};
+
+struct main0_patchIn
+{
+    patch_control_point<main0_in> gl_in;
+};
+
+[[ patch(quad, 0) ]] vertex main0_out main0(main0_patchIn patchIn [[stage_in]])
+{
+    main0_out out = {};
+    out.gl_Position.x = patchIn.gl_in[0].gl_ClipDistance_0;
+    out.gl_Position.y = patchIn.gl_in[1].gl_CullDistance_0;
+    out.gl_Position.z = patchIn.gl_in[0].gl_ClipDistance_1;
+    out.gl_Position.w = patchIn.gl_in[1].gl_CullDistance_1;
+    out.gl_Position += patchIn.gl_in[0].gl_Position;
+    out.gl_Position += patchIn.gl_in[1].gl_Position;
+    return out;
+}
+

shaders-msl-no-opt/tese/load-clip-cull.msl2.tese

@@ -0,0 +1,19 @@
+#version 450
+layout(quads) in;
+
+in gl_PerVertex
+{
+	float gl_ClipDistance[2];
+	float gl_CullDistance[3];
+	vec4 gl_Position;
+} gl_in[];
+
+void main()
+{
+	gl_Position.x = gl_in[0].gl_ClipDistance[0];
+	gl_Position.y = gl_in[1].gl_CullDistance[0];
+	gl_Position.z = gl_in[0].gl_ClipDistance[1];
+	gl_Position.w = gl_in[1].gl_CullDistance[1];
+	gl_Position += gl_in[0].gl_Position;
+	gl_Position += gl_in[1].gl_Position;
+}

spirv_glsl.cpp

@@ -8375,6 +8375,11 @@ void CompilerGLSL::access_chain_internal_append_index(std::string &expr, uint32_
 	expr += "]";
 }
 
+bool CompilerGLSL::access_chain_needs_stage_io_builtin_translation(uint32_t)
+{
+	return true;
+}
+
 string CompilerGLSL::access_chain_internal(uint32_t base, const uint32_t *indices, uint32_t count,
                                            AccessChainFlags flags, AccessChainMeta *meta)
 {
@@ -8584,7 +8589,7 @@ string CompilerGLSL::access_chain_internal(uint32_t base, const uint32_t *indice
 				SPIRV_CROSS_THROW("Member index is out of bounds!");
 
 			BuiltIn builtin;
-			if (is_member_builtin(*type, index, &builtin))
+			if (is_member_builtin(*type, index, &builtin) && access_chain_needs_stage_io_builtin_translation(base))
 			{
 				if (access_chain_is_arrayed)
 				{

spirv_glsl.hpp

@@ -673,6 +673,8 @@ protected:
 	std::string access_chain_internal(uint32_t base, const uint32_t *indices, uint32_t count, AccessChainFlags flags,
 	                                  AccessChainMeta *meta);
 
+	virtual bool access_chain_needs_stage_io_builtin_translation(uint32_t base);
+
 	virtual void prepare_access_chain_for_scalar_access(std::string &expr, const SPIRType &type,
 	                                                    spv::StorageClass storage, bool &is_packed);
 

spirv_msl.cpp

@@ -7354,6 +7354,21 @@ void CompilerMSL::prepare_access_chain_for_scalar_access(std::string &expr, cons
 	}
 }
 
+bool CompilerMSL::access_chain_needs_stage_io_builtin_translation(uint32_t base)
+{
+	auto *var = maybe_get_backing_variable(base);
+	if (!var || !is_tessellation_shader())
+		return true;
+
+	// We only need to rewrite builtin access chains when accessing flattened builtins like gl_ClipDistance_N.
+	// Avoid overriding it back to just gl_ClipDistance.
+	// This can only happen in scenarios where we cannot flatten/unflatten access chains, so, the only case
+	// where this triggers is evaluation shader inputs.
+	bool redirect_builtin = get_execution_model() == ExecutionModelTessellationEvaluation ?
+	                        var->storage == StorageClassOutput : false;
+	return redirect_builtin;
+}
+
 // Sets the interface member index for an access chain to a pull-model interpolant.
 void CompilerMSL::fix_up_interpolant_access_chain(const uint32_t *ops, uint32_t length)
 {

spirv_msl.hpp

@@ -951,6 +951,7 @@ protected:
 
 	void analyze_sampled_image_usage();
 
+	bool access_chain_needs_stage_io_builtin_translation(uint32_t base) override;
 	void prepare_access_chain_for_scalar_access(std::string &expr, const SPIRType &type, spv::StorageClass storage,
 	                                            bool &is_packed) override;
 	void fix_up_interpolant_access_chain(const uint32_t *ops, uint32_t length);