Geoshader support!

Manual.md

@@ -137,10 +137,10 @@ Reserves a new integer vector uniform to be preloaded with the specified constan
 ```
 Allocates a new output register, wires it to a certain output property and creates an alias for it that points to the allocated register. The following property names are supported:
 
-- `position` (or `pos`): In vertex shaders, this represents the position of the outputted vertex.
+- `position` (or `pos`): Represents the position of the outputted vertex.
 - `normalquat` (or `nquat`): Under investigation.
-- `color` (or `clr`): In vertex shaders, this represents the color of the outputted vertex. Its format is (R, G, B, xx) where R,G,B are values ranging from 0.0 to 1.0. The W component isn't used.
-- `texcoord0` (or `tcoord0`): In vertex shaders, this represents the texture coordinate that is fed to the Texture Unit 0. The Z and W components are not used.
+- `color` (or `clr`): Represents the color of the outputted vertex. Its format is (R, G, B, xx) where R,G,B are values ranging from 0.0 to 1.0. The W component isn't used.
+- `texcoord0` (or `tcoord0`): Represents the texture coordinate that is fed to the Texture Unit 0. The Z and W components are not used.
 - `texcoord0w` (or `tcoord0w`): Under investigation.
 - `texcoord1` (or `tcoord1`): As `texcoord0`, but for the Texture Unit 1.
 - `texcoord2` (or `tcoord2`): As `texcoord0`, but for the Texture Unit 2.
@@ -163,6 +163,8 @@ Syntax                            | Description
 --------------------------------- | -----------------------------------
 `nop`                             | No operation.
 `end`                             | Signals the end of the program.
+`emit`                            | (Geoshader-only) Emits a vertex configured by a prior `setemit`.
+`setemit vtxId, emitFlags         | (Geoshader-only) Configures a vertex for emission. The `emitFlags` parameter can be omitted.
 `add rDest, rSrc1, rSrc2`         |
 `dp3 rDest, rSrc1, rSrc2`         |
 `dp4 rDest, rSrc1, rSrc2`         |
@@ -217,3 +219,7 @@ Syntax                            | Description
 	- `flag1`: It tests a single flag.
 	- `flag1 && flag2`: It performs AND between the two flags. Optionally, a single `&` may be specified.
 	- `flag1 || flag2`: It performs OR between the two flags. Optionally, a single `|` may be specified.
+- `vtxId`: An integer ranging from 0 to 3 specifying the vertex ID used in geoshader vertex emission.
+- `emitFlags`: A space delimited combination of the following words:
+	- `primitive` (or `prim`): Specifies that after emitting the vertex, a primitive should also be emitted.
+	- `inv` (or `invert`): Specifies that the order of the vertices in the emitted primitive is inverted.

source/picasso.h

@@ -61,6 +61,7 @@ enum
 typedef std::vector<u32> outputBufType;
 typedef outputBufType::iterator outputBufIter;
 
+extern bool g_isGeoShader;
 extern outputBufType g_outputBuf;
 
 enum

source/picasso_assembler.cpp

@@ -7,6 +7,7 @@
 static const char* curFile = NULL;
 static int curLine = -1;
 
+bool g_isGeoShader = false;
 std::vector<u32> g_outputBuf;
 
 StackEntry g_stack[MAX_STACK];
@@ -561,9 +562,12 @@ static int parseReg(char* pos, int& outReg, int& outSw, int* idxType = NULL)
 	switch (*pos)
 	{
 		case 'o': // Output registers
-		case 'v': // Input attributes
 			if (outReg < 0x00 || outReg >= 0x08)
-				return throwError("invalid input/output register: %s(%d)\n", pos);
+				return throwError("invalid output register: %s(%d)\n", pos);
+			break;
+		case 'v': // Input attributes
+			if (outReg < 0x00 || outReg >= 0x0F)
+				return throwError("invalid input register: %s(%d)\n", pos);
 			break;
 		case 'r': // Temporary registers
 			outReg += 0x10;
@@ -805,6 +809,46 @@ DEF_COMMAND(formatmova)
 	return 0;
 }
 
+static inline int parseSetEmitFlags(char* flags, bool& isPrim, bool& isInv)
+{
+	isPrim = false;
+	isInv = false;
+	if (!flags)
+		return 0;
+
+	mystrtok_pos = flags;
+	while (char* flag = mystrtok_spc(NULL))
+	{
+		if (stricmp(flag, "prim")==0 || stricmp(flag, "primitive")==0)
+			isPrim = true;
+		else if (stricmp(flag, "inv")==0 || stricmp(flag, "invert")==0)
+			isInv = true;
+		else
+			throwError("unknown setemit flag: %s\n", flag);
+
+	}
+	return 0;
+}
+
+DEF_COMMAND(formatsetemit)
+{
+	NEXT_ARG(vtxIdStr);
+	NEXT_ARG_OPT(flagStr, NULL);
+	ENSURE_NO_MORE_ARGS();
+
+	ARG_TO_INT(vtxId, vtxIdStr, 0, 3);
+	bool isPrim, isInv;
+	safe_call(parseSetEmitFlags(flagStr, isPrim, isInv));
+
+#ifdef DEBUG
+	printf("%s:%02X vtx%d, %s, %s\n", cmdName, opcode, vtxId, isPrim?"true":"false", isInv=?"true":"false");
+#endif
+	BUF.push_back(FMT_OPCODE(opcode) | ((u32)isInv<<22) | ((u32)isPrim<<23) | (vtxId<<24));
+	g_isGeoShader = true;
+
+	return 0;
+}
+
 DEF_COMMAND(formatcall)
 {
 	NEXT_ARG(procName);
@@ -967,6 +1011,7 @@ static const cmdTableType cmdTable[] =
 {
 	DEC_COMMAND(NOP, format0),
 	DEC_COMMAND(END, format0),
+	DEC_COMMAND(EMIT, format0),
 
 	DEC_COMMAND(ADD, format1),
 	DEC_COMMAND(DP3, format1),
@@ -1009,6 +1054,8 @@ static const cmdTableType cmdTable[] =
 	DEC_COMMAND(MADI, format5i),
 	DEC_COMMAND(MAD, format5),
 
+	DEC_COMMAND(SETEMIT, formatsetemit),
+
 	{ NULL, NULL },
 };
 

source/picasso_frontend.cpp

@@ -105,7 +105,7 @@ int main(int argc, char* argv[])
 	
 	f.WriteWord(0x454C5644); // DVLE
 	f.WriteHword(0); // padding?
-	f.WriteHword(0); // Vertex shader
+	f.WriteHword(g_isGeoShader ? 1 : 0); // Shader type
 	f.WriteWord(mainIt->second.first); // offset to main
 	f.WriteWord(mainIt->second.first+mainIt->second.second); // offset to end of main
 	f.WriteWord(0); // ???
@@ -192,21 +192,22 @@ int main(int argc, char* argv[])
 
 		fprintf(f2, "// Generated by picasso\n");
 		fprintf(f2, "#pragma once\n");
+		const char* prefix = g_isGeoShader ? "GSH" : "VSH";
 		for (int i = 0; i < g_uniformCount; i ++)
 		{
 			Uniform& u = g_uniformTable[i];
 			if (u.type == UTYPE_FVEC)
-				fprintf(f2, "#define SHADER_FVEC_%s 0x%02X\n", u.name, u.pos-0x20);
+				fprintf(f2, "#define %s_FVEC_%s 0x%02X\n", prefix, u.name, u.pos-0x20);
 			else if (u.type == UTYPE_IVEC)
-				fprintf(f2, "#define SHADER_IVEC_%s 0x%02X\n", u.name, u.pos-0x80);
+				fprintf(f2, "#define %s_IVEC_%s 0x%02X\n", prefix, u.name, u.pos-0x80);
 			else if (u.type == UTYPE_BOOL)
 			{
 				if (u.size == 1)
-					fprintf(f2, "#define SHADER_FLAG_%s BIT(%d)\n", u.name, u.pos-0x88);
+					fprintf(f2, "#define %s_FLAG_%s BIT(%d)\n", prefix, u.name, u.pos-0x88);
 				else
-					fprintf(f2, "#define SHADER_FLAG_%s(_n) BIT(%d+(_n))\n", u.name, u.pos-0x88);
+					fprintf(f2, "#define %s_FLAG_%s(_n) BIT(%d+(_n))\n", prefix, u.name, u.pos-0x88);
 			}
-			fprintf(f2, "#define SHADER_ULEN_%s %d\n", u.name, u.size);
+			fprintf(f2, "#define %s_ULEN_%s %d\n", prefix, u.name, u.size);
 		}
 
 		fclose(f2);