prototype for spirv linking

funnyassets/shaders/funny_basic_pbr.shader

@@ -2,13 +2,18 @@
 
 COMMON
 {
-
+	struct RayPayload
+	{
+		float3 m_vOrigin;
+		float3 m_vDirection;
+		float3 m_vColor;
+	};
 }
 CALLABLE
 {
 	#include "textures.hlsl"
-	float4 brdfMain()
+	void brdfMain( inout RayPayload payload )
 	{
-		return float4(1);
+		payload.m_vColor = float3(1.0,0.0,1.0);
 	}
 }

funnyassets/shaders/funny_error.shader

@@ -1,9 +1,19 @@
 #include "macros.hlsl"
 
+COMMON
+{
+	struct RayPayload
+	{
+		float3 m_vOrigin;
+		float3 m_vDirection;
+		float3 m_vColor;
+	};
+}
 CALLABLE
 {
-	float4 CallableMain()
+	#include "textures.hlsl"
+	void CallableMain( inout RayPayload payload )
 	{
-		return float4(1);
+		payload.m_vColor = float3(1.0,0.0,1.0);
 	}
 }

funnyassets/shaders/macros.hlsl

@@ -27,26 +27,29 @@
 #define CALLABLE namespace CallableShader_DO_NOT_USE
 #endif
 
-float4 Test()
+/* 
+ * we kinda want to lookup OpExecuteCallableKHR inside of runtime linker
+ * and replace _CallShader with our own
+ * but stupid slangc can't do noinline 
+ * so we have to do id += 1 and id -= 1
+ * overall trick 5/10
+ */
+[noinline]
+void _CallShader2<Payload>( uint32_t id, inout Payload data )
 {
-	return float4(1,0.5,1,1);
+	id -= 1;
+	spirv_asm
+	{
+		OpExecuteCallableKHR $id $data;
+	};
 }
 
-#ifndef USE_CALLABLE_SHADERS
-#define USE_CALLABLE_SHADERS
-#endif
-
-struct RunShaderResult_t<A>
-{
-	A val;
-
-};
-
 [noinline]
-[builtin]
-RunShaderResult_t<A> RunShader<A, B>( uint32_t id, B data ) where optional B
+void _CallShader<Payload>( uint32_t id, inout Payload data )
 {
-	return {};
+	id += 1;
+	_CallShader2(id, data);
 }
+#define CallShader _CallShader
 
 #endif

funnyassets/shaders/mesh_raster.shader

@@ -2,8 +2,6 @@
 #include "funny_shared.hlsl"
 
 
-USE_CALLABLE_SHADERS;
-
 COMMON {
 	cbuffer CameraInfo 
 	{

funnyassets/shaders/mesh_trace.shader

@@ -3,12 +3,12 @@
 
 struct MeshPayload
 {
-	float3 m_vColor;
 }
 struct RayPayload
 {
 	float3 m_vOrigin;
 	float3 m_vDirection;
+	float3 m_vColor;
 };
 
 RAY
@@ -23,9 +23,8 @@ RAY
 	void rayMain()
 	{
 		RayPayload p = {};
-		RunShaderResult_t<MeshPayload> m = RunShader<MeshPayload, RayPayload>(0, p);
-		m.val.m_vColor;
-		printf("%f\n", m.val.m_vColor.x);
+		CallShader<RayPayload>(0, p);
+		printf("%f\n", p.m_vColor.x);
 		
 	/*
 		uint2 pixel = DispatchRaysIndex().xy;

materialsystem/compiledshader.cpp

@@ -28,12 +28,8 @@ ShaderObject_t *CCompiledShader::AllocateShader()
 
 ShaderObject_t *CCompiledShader::FindShaderObject( EShaderBackend eBackend, EShaderStage eStage )
 {
-	V_printf("%i\n", m_objects.GetSize());
-	V_printf("%i\n", m_lumps.GetSize());
 	for ( auto &o: m_objects )
 	{
-		V_printf("%i\n", o.m_eBackend);
-		V_printf("%i\n", o.m_eStage);
 		if ( o.m_eBackend != eBackend )
 			continue;
 

materialsystem/vulkan/rtlinker.cpp

@@ -36,6 +36,12 @@ void CVkShaderLinker::Build()
 	uint32_t last = 0;
 	uint32_t current = 0;
 
+
+	CUtlVector<uint32_t> callers = {};
+	mspv_array(mspv_function) functions_orig;
+	V_memcpy(&functions_orig, &mod->functions, sizeof(functions_orig));
+	mod->functions = {};
+
 	for ( int i = 0; i < mod->types.count; i++ )
 	{
 		SET_LAST(mod->types.data[i].id);
@@ -50,13 +56,20 @@ void CVkShaderLinker::Build()
 		SET_LAST(mod->variables.data[i].result);
 		SET_LAST(mod->variables.data[i].resulttype);
 	}
-	for ( int i = 0; i < mod->functions.count; i++ )
+	for ( int i = 0; i < functions_orig.count; i++ )
 	{
-		SET_LAST(mod->functions.data[i].result);
-		mspv_function &f = mod->functions.data[i];
+		SET_LAST(functions_orig.data[i].result);
+		mspv_function &f = functions_orig.data[i];
+		bool bIsCaller = false;
 		for ( int u = 0; u < f.instructions.len; )
 		{
 			SpvOp op = (SpvOp)(f.instructions.data[u]&0xFFFF);
+			if (op == SpvOpExecuteCallableKHR)
+			{
+				bIsCaller = true;
+				callers.AppendTail(i);
+				V_printf("found caller: %i %u\n", i, f.result);
+			}
 			uint32_t uOpLen = f.instructions.data[u]>>16;
 			uint32_t uOpCount = SpvGetOperandCount(op);
 			if (uOpCount == 0)
@@ -90,13 +103,16 @@ void CVkShaderLinker::Build()
 			}
 			u+=uOpLen;
 			V_free(peOps);
-
+			V_free(peFlags);
 		}
+		if (!bIsCaller)
+			mspv_array_push(mod->functions, f);
 	}
 	current = last;
 
 	/* now we can combine shaders*/
 	CUtlVector<uint32_t> functions = {};
+	CUtlVector<mspv_function> f = {};
 	for (auto &shader: m_shaders)
 	{
 		mspv_module *s = mspv_read_module(shader.m_size, shader.m_data);
@@ -117,6 +133,14 @@ void CVkShaderLinker::Build()
 		for ( int i = 0; i < s->entry_points.count; i++ )
 		{
 			functions.AppendTail(s->entry_points.data[i].id+current);
+			for ( uint32_t u = 0; u < s->functions.count; u++ )
+			{
+				if (s->functions.data[u].result != s->entry_points.data[i].id)
+					continue;
+				f.AppendTail(s->functions.data[u]);
+				break;
+
+			}
 			SET_LAST(s->entry_points.data[i].id+current);
 		}
 
@@ -222,6 +246,52 @@ void CVkShaderLinker::Build()
 		}
 		current = last;
 	}
+	/* generate trace call */
+	V_printf("functions %i\n", functions.GetSize());
+	uint32_t n = 0;
+	for ( auto &c: callers )
+	{
+		CUtlVector<uint32_t> ops = {};
+		mspv_function fn = functions_orig.data[c];
+		mspv_data_view dv = fn.instructions;
+		uint32_t u;
+		CUtlVector<uint32_t> inputs = {};
+		
+		u = 0;
+		while(u < fn.instructions.len)
+		{	
+			SpvOp op = (SpvOp)(dv.data[u]&0xFFFF);
+			uint32_t uOpLen = dv.data[u]>>16;
+			if (op != SpvOpFunctionParameter)
+			{
+				break;
+			}
+			ops.AppendTail(&dv.data[u], uOpLen);
+			inputs.AppendTail(dv.data[u+2]);
+			u += uOpLen;
+		}
+		ops.AppendTail(mspv_make_op(SpvOpLabel, 1));
+		ops.AppendTail(current++);
+
+		ops.AppendTail(mspv_make_op(SpvOpFunctionCall, 5));
+		ops.AppendTail(f[n].resulttype);
+		ops.AppendTail(current++);
+		ops.AppendTail(f[n].result);
+		ops.AppendTail(inputs[0]);
+		ops.AppendTail(inputs[1]);
+
+		ops.AppendTail(mspv_make_op(SpvOpReturn, 0));
+
+
+
+		fn.instructions.data = (uint32_t*)V_malloc(ops.GetSize()*sizeof(uint32_t));
+		V_memcpy(fn.instructions.data, ops.GetData(), ops.GetSize()*sizeof(uint32_t));
+		fn.instructions.len = ops.GetSize();
+		mspv_array_push(mod->functions, fn);
+
+		n++;
+
+	}
 
 	mspv_spv spv = mspv_write_module(mod);
 	mspv_close_module(mod);

minispv/minispv.h

@@ -353,6 +353,11 @@ static inline void mspv_write_op( mspv_spv *as, SpvOp op, uint16_t ops )
 		mspv_array_push((*as), op | (ops+1) << 16);
 };
 
+static inline uint32_t mspv_make_op( SpvOp op, uint16_t ops )
+{
+	return op | (ops+1) << 16;
+};
+
 static inline void mspv_write_string( mspv_spv *as, mspv_string_view s )
 {
 	uint32_t u;