r300/compiler: Introduce aluresult register for branch operation support

Signed-off-by: Nicolai Hähnle <nhaehnle@gmail.com>

7 files changed, 154 insertions, 15 deletions

diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
index 590201a9ba..614c2e3d24 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
@@ -88,11 +88,12 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 	if (c->is_r500) {
 		struct radeon_program_transformation transformations[] = {
 			{ &r500_transform_TEX, c },
+			{ &r500_transform_IF, 0 },
 			{ &radeonTransformALU, 0 },
 			{ &radeonTransformDeriv, 0 },
 			{ &radeonTransformTrigScale, 0 }
 		};
-		radeonLocalTransform(&c->Base, 4, transformations);
+		radeonLocalTransform(&c->Base, 5, transformations);
 
 		c->Base.SwizzleCaps = &r500_swizzle_caps;
 	} else {
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
index 971465e359..39f2445bd4 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
@@ -169,6 +169,34 @@ int r500_transform_TEX(
 	return 1;
 }
 
+/**
+ * Rewrite IF instructions to use the ALU result special register.
+ */
+int r500_transform_IF(
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
+	void* data)
+{
+	if (inst->I.Opcode != RC_OPCODE_IF)
+		return 0;
+
+	struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+	inst_mov->I.Opcode = RC_OPCODE_MOV;
+	inst_mov->I.DstReg.WriteMask = 0;
+	inst_mov->I.WriteALUResult = RC_ALURESULT_W;
+	inst_mov->I.ALUResultCompare = RC_COMPARE_FUNC_NOTEQUAL;
+	inst_mov->I.SrcReg[0] = inst->I.SrcReg[0];
+	inst_mov->I.SrcReg[0].Swizzle = combine_swizzles4(inst_mov->I.SrcReg[0].Swizzle,
+			RC_SWIZZLE_UNUSED, RC_SWIZZLE_UNUSED, RC_SWIZZLE_UNUSED, RC_SWIZZLE_X);
+
+	inst->I.SrcReg[0].File = RC_FILE_SPECIAL;
+	inst->I.SrcReg[0].Index = RC_SPECIAL_ALU_RESULT;
+	inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW;
+	inst->I.SrcReg[0].Negate = 0;
+
+	return 1;
+}
+
 static int r500_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
 {
 	unsigned int relevant;
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
index 92ac75d5fd..0918cdf518 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
@@ -47,4 +47,9 @@ extern int r500_transform_TEX(
 	struct rc_instruction * inst,
 	void* data);
 
+extern int r500_transform_IF(
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
+	void* data);
+
 #endif
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
index 95af6fd411..2ae3c56689 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
@@ -34,10 +34,12 @@ struct updatemask_state {
 	unsigned char Output[RC_REGISTER_MAX_INDEX];
 	unsigned char Temporary[RC_REGISTER_MAX_INDEX];
 	unsigned char Address;
+	unsigned char Special[RC_NUM_SPECIAL_REGISTERS];
 };
 
 struct instruction_state {
-	unsigned char WriteMask;
+	unsigned char WriteMask:4;
+	unsigned char WriteALUResult:1;
 	unsigned char SrcReg[3];
 };
 
@@ -70,6 +72,9 @@ static void or_updatemasks(
 		dst->Temporary[i] = a->Temporary[i] | b->Temporary[i];
 	}
 
+	for(unsigned int i = 0; i < RC_NUM_SPECIAL_REGISTERS; ++i)
+		dst->Special[i] = a->Special[i] | b->Special[i];
+
 	dst->Address = a->Address | b->Address;
 }
 
@@ -108,6 +113,13 @@ static unsigned char * get_used_ptr(struct deadcode_state *s, rc_register_file f
 			return &s->R.Temporary[index];
 	} else if (file == RC_FILE_ADDRESS) {
 		return &s->R.Address;
+	} else if (file == RC_FILE_SPECIAL) {
+		if (index >= RC_NUM_SPECIAL_REGISTERS) {
+			rc_error(s->C, "%s: special file index %i out of bounds\n", __FUNCTION__, index);
+			return 0;
+		}
+
+		return &s->R.Special[index];
 	}
 
 	return 0;
@@ -136,6 +148,19 @@ static void update_instruction(struct deadcode_state * s, struct rc_instruction
 
 	insts->WriteMask |= usedmask;
 
+	if (inst->I.WriteALUResult) {
+		unsigned char * pused = get_used_ptr(s, RC_FILE_SPECIAL, RC_SPECIAL_ALU_RESULT);
+		if (pused && *pused) {
+			if (inst->I.WriteALUResult == RC_ALURESULT_X)
+				usedmask |= RC_MASK_X;
+			else if (inst->I.WriteALUResult == RC_ALURESULT_W)
+				usedmask |= RC_MASK_W;
+
+			*pused = 0;
+			insts->WriteALUResult = 1;
+		}
+	}
+
 	unsigned int srcmasks[3];
 	rc_compute_sources_for_writemask(opcode, usedmask, srcmasks);
 
@@ -225,21 +250,38 @@ void rc_dataflow_deadcode(struct radeon_compiler * c, rc_dataflow_mark_outputs_f
 	for(struct rc_instruction * inst = c->Program.Instructions.Next;
 	    inst != &c->Program.Instructions;
 	    inst = inst->Next, ++ip) {
-		const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode);
+		const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode);\
+		int dead = 1;
+
+		if (!opcode->HasDstReg) {
+			dead = 0;
+		} else {
+			inst->I.DstReg.WriteMask = s.Instructions[ip].WriteMask;
+			if (s.Instructions[ip].WriteMask)
+				dead = 0;
+
+			if (s.Instructions[ip].WriteALUResult)
+				dead = 0;
+			else
+				inst->I.WriteALUResult = RC_ALURESULT_NONE;
+		}
 
-		if (opcode->HasDstReg) {
-			if (s.Instructions[ip].WriteMask) {
-				inst->I.DstReg.WriteMask = s.Instructions[ip].WriteMask;
-			} else {
-				struct rc_instruction * todelete = inst;
-				inst = inst->Prev;
-				rc_remove_instruction(todelete);
-				continue;
-			}
+		if (dead) {
+			struct rc_instruction * todelete = inst;
+			inst = inst->Prev;
+			rc_remove_instruction(todelete);
+			continue;
 		}
 
 		unsigned int srcmasks[3];
-		rc_compute_sources_for_writemask(opcode, s.Instructions[ip].WriteMask, srcmasks);
+		unsigned int usemask = s.Instructions[ip].WriteMask;
+
+		if (inst->I.WriteALUResult == RC_ALURESULT_X)
+			usemask |= RC_MASK_X;
+		else if (inst->I.WriteALUResult == RC_ALURESULT_W)
+			usemask |= RC_MASK_W;
+
+		rc_compute_sources_for_writemask(opcode, usemask, srcmasks);
 
 		for(unsigned int src = 0; src < 3; ++src) {
 			for(unsigned int chan = 0; chan < 4; ++chan) {
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.h b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
index efa2b0dfe3..071b0a0ca9 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
@@ -87,6 +87,14 @@ struct rc_sub_instruction {
 	rc_saturate_mode SaturateMode:2;
 
 	/**
+	 * Writing to the special register RC_SPECIAL_ALU_RESULT
+	 */
+	/*@{*/
+	rc_write_aluresult WriteALUResult:2;
+	rc_compare_func ALUResultCompare:3;
+	/*@}*/
+
+	/**
 	 * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions.
 	 */
 	/*@{*/
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h
index 69994f9880..7c0d6720b1 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h
@@ -74,9 +74,22 @@ typedef enum {
 	/**
 	 * Indicates a constant from the \ref rc_constant_list .
 	 */
-	RC_FILE_CONSTANT
+	RC_FILE_CONSTANT,
+
+	/**
+	 * Indicates a special register, see RC_SPECIAL_xxx.
+	 */
+	RC_FILE_SPECIAL
 } rc_register_file;
 
+enum {
+	/** R500 fragment program ALU result "register" */
+	RC_SPECIAL_ALU_RESULT = 0,
+
+	/** Must be last */
+	RC_NUM_SPECIAL_REGISTERS
+};
+
 #define RC_REGISTER_INDEX_BITS 10
 #define RC_REGISTER_MAX_INDEX (1 << RC_REGISTER_INDEX_BITS)
 
@@ -125,4 +138,10 @@ typedef enum {
 #define RC_MASK_XYZW (RC_MASK_X|RC_MASK_Y|RC_MASK_Z|RC_MASK_W)
 /*@}*/
 
+typedef enum {
+	RC_ALURESULT_NONE = 0,
+	RC_ALURESULT_X,
+	RC_ALURESULT_W
+} rc_write_aluresult;
+
 #endif /* RADEON_PROGRAM_CONSTANTS_H */
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c
index 0485286451..6645d7cacb 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c
@@ -38,10 +38,36 @@ static const char * textarget_to_string(rc_texture_target target)
 	}
 }
 
+static void rc_print_comparefunc(FILE * f, const char * lhs, rc_compare_func func, const char * rhs)
+{
+	if (func == RC_COMPARE_FUNC_NEVER) {
+		fprintf(f, "false");
+	} else if (func == RC_COMPARE_FUNC_ALWAYS) {
+		fprintf(f, "true");
+	} else {
+		const char * op;
+		switch(func) {
+		case RC_COMPARE_FUNC_LESS: op = "<"; break;
+		case RC_COMPARE_FUNC_EQUAL: op = "=="; break;
+		case RC_COMPARE_FUNC_LEQUAL: op = "<="; break;
+		case RC_COMPARE_FUNC_GREATER: op = ">"; break;
+		case RC_COMPARE_FUNC_NOTEQUAL: op = "!="; break;
+		case RC_COMPARE_FUNC_GEQUAL: op = ">="; break;
+		default: op = "???"; break;
+		}
+		fprintf(f, "%s %s %s", lhs, op, rhs);
+	}
+}
+
 static void rc_print_register(FILE * f, rc_register_file file, int index, unsigned int reladdr)
 {
 	if (file == RC_FILE_NONE) {
 		fprintf(f, "none");
+	} else if (file == RC_FILE_SPECIAL) {
+		switch(index) {
+		case RC_SPECIAL_ALU_RESULT: fprintf(f, "aluresult"); break;
+		default: fprintf(f, "special[%i]", index); break;
+		}
 	} else {
 		const char * filename;
 		switch(file) {
@@ -151,7 +177,17 @@ static void rc_print_instruction(FILE * f, struct rc_instruction * inst)
 			inst->I.TexSrcUnit);
 	}
 
-	fprintf(f, ";\n");
+	fprintf(f, ";");
+
+	if (inst->I.WriteALUResult) {
+		fprintf(f, " [aluresult = (");
+		rc_print_comparefunc(f,
+			(inst->I.WriteALUResult == RC_ALURESULT_X) ? "x" : "w",
+			inst->I.ALUResultCompare, "0");
+		fprintf(f, ")]");
+	}
+
+	fprintf(f, "\n");
 }
 
 /**