r300/compiler: add new compiler parameter max_alu_insts

12 files changed, 38 insertions, 35 deletions

diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index 9845e546109..2b9861cae38 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -386,6 +386,7 @@ static void r300_translate_fragment_shader(
     compiler.state = shader->compare_state;
     compiler.Base.is_r500 = r300->screen->caps.is_r500;
     compiler.Base.max_temp_regs = compiler.Base.is_r500 ? 128 : 32;
+    compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64;
     compiler.Base.remove_unused_constants = TRUE;
     compiler.AllocateHwInputs = &allocate_hardware_inputs;
     compiler.UserData = &shader->inputs;
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 5f8dbb28d0c..29569d92f59 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -206,6 +206,7 @@ void r300_translate_vertex_shader(struct r300_context *r300,
     compiler.UserData = vs;
     compiler.Base.is_r500 = r300->screen->caps.is_r500;
     compiler.Base.max_temp_regs = 32;
+    compiler.Base.max_alu_insts = r300->screen->caps.is_r500 ? 1024 : 256;
     compiler.Base.remove_unused_constants = TRUE;
 
     if (compiler.Base.Debug) {
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
index b27a683c397..ba5461bef48 100644
--- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
@@ -135,7 +135,7 @@ static int emit_alu(struct r300_emit_state * emit, struct rc_pair_instruction* i
 {
 	PROG_CODE;
 
-	if (code->alu.length >= R300_PFS_MAX_ALU_INST) {
+	if (code->alu.length >= c->Base.max_alu_insts) {
 		error("Too many ALU instructions");
 		return 0;
 	}
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
index f82d0f1e62d..96c5d6fda9f 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
@@ -107,12 +107,11 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 
 	debug_program_log(c, "before compilation");
 
-	if (c->Base.is_r500){
-		rc_unroll_loops(&c->Base, R500_PFS_MAX_INST);
+	if (c->Base.is_r500) {
+		rc_unroll_loops(&c->Base);
 		debug_program_log(c, "after unroll loops");
-	}
-	else{
-		rc_transform_loops(&c->Base, -1);
+	} else {
+		rc_transform_loops(&c->Base);
 		debug_program_log(c, "after transform loops");
 
 		rc_emulate_branches(&c->Base);
@@ -165,7 +164,7 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 	debug_program_log(c, "after deadcode");
 
 	if (!c->Base.is_r500) {
-		rc_emulate_loops(&c->Base, R300_PFS_MAX_ALU_INST);
+		rc_emulate_loops(&c->Base);
 		debug_program_log(c, "after emulate loops");
 	}
 
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
index b5186809bc3..29c6c869ff7 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -985,10 +985,7 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler *c)
 
 	debug_program_log(c, "before compilation");
 
-	if (c->Base.is_r500)
-		rc_transform_loops(&c->Base, R500_VS_MAX_ALU);
-	else
-		rc_transform_loops(&c->Base, R300_VS_MAX_ALU);
+	rc_transform_loops(&c->Base);
 	if (c->Base.Error)
 		return;
 
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
index 5612a4e5ccc..8a6aafe5af1 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
@@ -229,7 +229,7 @@ static void emit_paired(struct r300_fragment_program_compiler *c, struct rc_pair
 {
 	PROG_CODE;
 
-	if (code->inst_end >= 511) {
+	if (code->inst_end >= c->Base.max_alu_insts-1) {
 		error("emit_alu: Too many instructions");
 		return;
 	}
@@ -322,7 +322,7 @@ static int emit_tex(struct r300_fragment_program_compiler *c, struct rc_sub_inst
 {
 	PROG_CODE;
 
-	if (code->inst_end >= 511) {
+	if (code->inst_end >= c->Base.max_alu_insts-1) {
 		error("emit_tex: Too many instructions");
 		return 0;
 	}
@@ -370,7 +370,7 @@ static int emit_tex(struct r300_fragment_program_compiler *c, struct rc_sub_inst
 
 static void emit_flowcontrol(struct emit_state * s, struct rc_instruction * inst)
 {
-	if (s->Code->inst_end >= 511) {
+	if (s->Code->inst_end >= s->C->max_alu_insts-1) {
 		rc_error(s->C, "emit_tex: Too many instructions");
 		return;
 	}
@@ -577,7 +577,7 @@ void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compi
 		}
 	}
 
-	if (code->max_temp_idx >= 128)
+	if (code->max_temp_idx >= compiler->Base.max_temp_regs)
 		rc_error(&compiler->Base, "Too many hardware temporaries used");
 
 	if (compiler->Base.Error)
@@ -587,7 +587,7 @@ void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compi
 	    (code->inst[code->inst_end].inst0 & R500_INST_TYPE_MASK) != R500_INST_TYPE_OUT) {
 		/* This may happen when dead-code elimination is disabled or
 		 * when most of the fragment program logic is leading to a KIL */
-		if (code->inst_end >= 511) {
+		if (code->inst_end >= compiler->Base.max_alu_insts-1) {
 			rc_error(&compiler->Base, "Introducing fake OUT: Too many instructions");
 			return;
 		}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
index a61c005274a..1833fcebcac 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
@@ -42,6 +42,7 @@ struct radeon_compiler {
 	/* Hardware specification. */
 	unsigned is_r500:1;
 	unsigned max_temp_regs;
+	int max_alu_insts;
 
 	/* Whether to remove unused constants and empty holes in constant space. */
 	unsigned remove_unused_constants:1;
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 dbb41767633..d13b2be42db 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
@@ -78,12 +78,12 @@ static int src_reg_is_immediate(struct rc_src_register * src,
 }
 
 static unsigned int loop_max_possible_iterations(struct radeon_compiler *c,
-			struct loop_info * loop, unsigned int prog_inst_limit)
+			struct loop_info * loop)
 {
 	unsigned int total_i = rc_recompute_ips(c);
 	unsigned int loop_i = (loop->EndLoop->IP - loop->BeginLoop->IP) - 1;
 	/* +1 because the program already has one iteration of the loop. */
-	return 1 + ((prog_inst_limit - total_i) / loop_i);
+	return 1 + ((c->max_alu_insts - total_i) / loop_i);
 }
 
 static void unroll_loop(struct radeon_compiler * c, struct loop_info * loop,
@@ -187,11 +187,10 @@ 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
+ * If c->max_alu_inst 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)
+static int try_unroll_loop(struct radeon_compiler * c, struct loop_info * loop)
 {
 	int end_loops;
 	int iterations;
@@ -300,9 +299,8 @@ static int try_unroll_loop(struct radeon_compiler * c, struct loop_info * loop,
 		return 0;
 	}
 
-	if (prog_inst_limit > 0
-		&& iterations > loop_max_possible_iterations(c, loop,
-							prog_inst_limit)) {
+	if (c->max_alu_insts > 0
+		&& iterations > loop_max_possible_iterations(c, loop)) {
 		return 0;
 	}
 
@@ -436,7 +434,7 @@ static int transform_loop(struct emulate_loop_state * s,
 	if (!build_loop_info(s->C, loop, inst))
 		return 0;
 
-	if(try_unroll_loop(s->C, loop, s->prog_inst_limit)){
+	if(try_unroll_loop(s->C, loop)){
 		return 1;
 	}
 
@@ -472,14 +470,13 @@ static int transform_loop(struct emulate_loop_state * s,
 	return 1;
 }
 
-void rc_transform_loops(struct radeon_compiler *c, int prog_inst_limit)
+void rc_transform_loops(struct radeon_compiler *c)
 {
 	struct emulate_loop_state * s = &c->loop_state;
 	struct rc_instruction * ptr;
 
 	memset(s, 0, sizeof(struct emulate_loop_state));
 	s->C = c;
-	s->prog_inst_limit = prog_inst_limit;
 	for(ptr = s->C->Program.Instructions.Next;
 			ptr != &s->C->Program.Instructions; ptr = ptr->Next) {
 		if(ptr->Type == RC_INSTRUCTION_NORMAL &&
@@ -490,7 +487,7 @@ void rc_transform_loops(struct radeon_compiler *c, int prog_inst_limit)
 	}
 }
 
-void rc_unroll_loops(struct radeon_compiler *c, int prog_inst_limit)
+void rc_unroll_loops(struct radeon_compiler *c)
 {
 	struct rc_instruction * inst;
 	struct loop_info loop;
@@ -500,13 +497,13 @@ void rc_unroll_loops(struct radeon_compiler *c, int prog_inst_limit)
 
 		if (inst->U.I.Opcode == RC_OPCODE_BGNLOOP) {
 			if (build_loop_info(c, &loop, inst)) {
-				try_unroll_loop(c, &loop, prog_inst_limit);
+				try_unroll_loop(c, &loop);
 			}
 		}
 	}
 }
 
-void rc_emulate_loops(struct radeon_compiler *c, int prog_inst_limit)
+void rc_emulate_loops(struct radeon_compiler *c)
 {
 	struct emulate_loop_state * s = &c->loop_state;
 	int i;
@@ -518,7 +515,7 @@ void rc_emulate_loops(struct radeon_compiler *c, int prog_inst_limit)
 			continue;
 		}
 		unsigned int iterations = loop_max_possible_iterations(
-					s->C, &s->Loops[i], prog_inst_limit);
+					s->C, &s->Loops[i]);
 		unroll_loop(s->C, &s->Loops[i], iterations);
 	}
 }
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 43af8ee480e..caf1c576588 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.h
@@ -21,13 +21,12 @@ struct emulate_loop_state {
 	struct loop_info * Loops;
 	unsigned int LoopCount;
 	unsigned int LoopReserved;
-	int prog_inst_limit;
 };
 
-void rc_transform_loops(struct radeon_compiler *c, int prog_inst_limit);
+void rc_transform_loops(struct radeon_compiler *c);
 
-void rc_unroll_loops(struct radeon_compiler * c, int prog_inst_limit);
+void rc_unroll_loops(struct radeon_compiler * c);
 
-void rc_emulate_loops(struct radeon_compiler * c, int prog_inst_limit);
+void rc_emulate_loops(struct radeon_compiler * c);
 
 #endif /* RADEON_EMULATE_LOOPS_H */
diff --git a/src/mesa/drivers/dri/r300/r300_blit.c b/src/mesa/drivers/dri/r300/r300_blit.c
index d2c25fb9cd4..91d715a7169 100644
--- a/src/mesa/drivers/dri/r300/r300_blit.c
+++ b/src/mesa/drivers/dri/r300/r300_blit.c
@@ -88,6 +88,9 @@ static void create_vertex_program(struct r300_context *r300)
     compiler.RequiredOutputs = compiler.Base.Program.OutputsWritten = (1 << VERT_RESULT_HPOS) | (1 << VERT_RESULT_TEX0);
     compiler.SetHwInputOutput = vp_ins_outs;
     compiler.code = &r300->blit.vp_code;
+    compiler.Base.is_r500 = r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515;
+    compiler.Base.max_temp_regs = 32;
+    compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 1024 : 256;
 
     r3xx_compile_vertex_program(&compiler);
 }
@@ -120,6 +123,7 @@ static void create_fragment_program(struct r300_context *r300)
     compiler.enable_shadow_ambient = GL_TRUE;
     compiler.Base.is_r500 = (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515);
     compiler.Base.max_temp_regs = (compiler.Base.is_r500) ? 128 : 32;
+    compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64;
     compiler.code = &r300->blit.fp_code;
     compiler.AllocateHwInputs = fp_allocate_hw_inputs;
 
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
index 7b6521c7480..d6d41b4e42c 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_common.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
@@ -221,6 +221,7 @@ static void translate_fragment_program(GLcontext *ctx, struct r300_fragment_prog
 	compiler.enable_shadow_ambient = GL_TRUE;
 	compiler.Base.is_r500 = (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) ? GL_TRUE : GL_FALSE;
 	compiler.Base.max_temp_regs = (compiler.Base.is_r500) ? 128 : 32;
+	compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64;
 	compiler.OutputDepth = FRAG_RESULT_DEPTH;
 	memset(compiler.OutputColor, 0, 4 * sizeof(unsigned));
 	compiler.OutputColor[0] = FRAG_RESULT_COLOR;
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index 67d8b2b3286..948517bd807 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -244,6 +244,9 @@ static struct r300_vertex_program *build_program(GLcontext *ctx,
 	compiler.code = &vp->code;
 	compiler.RequiredOutputs = compute_required_outputs(vp->Base, vp->key.FpReads);
 	compiler.SetHwInputOutput = &t_inputs_outputs;
+	compiler.Base.is_r500 = R300_CONTEXT(ctx)->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515;
+	compiler.Base.max_temp_regs = 32;
+	compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 1024 : 256;
 
 	if (compiler.Base.Debug) {
 		fprintf(stderr, "Initial vertex program:\n");