r300/compiler: Always unroll loops when doing loop emulation.

4 files changed, 14 insertions, 11 deletions

diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
index 070939497ce..c6246a81a24 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
@@ -113,7 +113,7 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 		debug_program_log(c, "after unroll loops");
 	}
 	else{
-		rc_transform_loops(&c->Base, &loop_state, R300_PFS_MAX_ALU_INST);
+		rc_transform_loops(&c->Base, &loop_state);
 		debug_program_log(c, "after transform loops");
 
 		rc_emulate_branches(&c->Base);
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
index fe34ff67cdc..e940fedec20 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -645,11 +645,11 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler)
 	 * on all chipsets. */
 
 	if (compiler->Base.is_r500){
-		rc_transform_loops(&compiler->Base, &loop_state, R500_VS_MAX_ALU);
+		rc_transform_loops(&compiler->Base, &loop_state);
 		rc_emulate_loops(&loop_state, R500_VS_MAX_ALU);
 	} else {
+		rc_transform_loops(&compiler->Base, &loop_state);
 		rc_emulate_loops(&loop_state, R300_VS_MAX_ALU);
-		rc_transform_loops(&compiler->Base, &loop_state, R300_VS_MAX_ALU);
 	}
 	debug_program_log(compiler, "after emulate loops");
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
index 94e3e5f4f51..24c3ae57b6e 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
@@ -186,6 +186,10 @@ static void get_incr_amount(void * data, struct rc_instruction * inst,
 	}
 }
 
+/**
+ * If prog_inst_limit is -1, then all eligible loops will be unrolled regardless
+ * of how many iterations they have.
+ */
 static int try_unroll_loop(struct radeon_compiler * c, struct loop_info * loop,
 						unsigned int prog_inst_limit)
 {
@@ -296,7 +300,8 @@ static int try_unroll_loop(struct radeon_compiler * c, struct loop_info * loop,
 		return 0;
 	}
 
-	if (iterations > loop_max_possible_iterations(c, loop,
+	if (prog_inst_limit > 0
+		&& iterations > loop_max_possible_iterations(c, loop,
 							prog_inst_limit)) {
 		return 0;
 	}
@@ -422,8 +427,7 @@ static int build_loop_info(struct radeon_compiler * c, struct loop_info * loop,
  * 	   Null if there is an error.
  */
 static struct rc_instruction * transform_loop(struct emulate_loop_state * s,
-						struct rc_instruction * inst,
-						int prog_inst_limit)
+						struct rc_instruction * inst)
 {
 	struct loop_info * loop;
 
@@ -435,7 +439,7 @@ static struct rc_instruction * transform_loop(struct emulate_loop_state * s,
 	if (!build_loop_info(s->C, loop, inst))
 		return NULL;
 
-	if(try_unroll_loop(s->C, loop, prog_inst_limit)){
+	if(try_unroll_loop(s->C, loop, -1)){
 		return loop->BeginLoop->Next;
 	}
 
@@ -472,8 +476,7 @@ static struct rc_instruction * transform_loop(struct emulate_loop_state * s,
 }
 
 void rc_transform_loops(struct radeon_compiler *c,
-					struct emulate_loop_state * s,
-					int prog_inst_limit)
+						struct emulate_loop_state * s)
 {
 	struct rc_instruction * ptr;
 
@@ -483,7 +486,7 @@ void rc_transform_loops(struct radeon_compiler *c,
 	while(ptr != &s->C->Program.Instructions) {
 		if(ptr->Type == RC_INSTRUCTION_NORMAL &&
 					ptr->U.I.Opcode == RC_OPCODE_BGNLOOP){
-			ptr = transform_loop(s, ptr, prog_inst_limit);
+			ptr = transform_loop(s, ptr);
 			if(!ptr){
 				return;
 			}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h
index 339527ba3bd..86d91ef14bd 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h
@@ -24,7 +24,7 @@ struct emulate_loop_state {
 };
 
 void rc_transform_loops(struct radeon_compiler *c,
-			struct emulate_loop_state * s, int prog_inst_limit);
+						struct emulate_loop_state * s);
 
 void rc_unroll_loops(struct radeon_compiler * c, int prog_inst_limit);