r600g: optimize away CF_INST_POP

If last instruction is an CF_INST_ALU we don't need to emit an
additional CF_INST_POP for stack clean up after an IF ELSE ENDIF.

2 files changed, 27 insertions, 3 deletions

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 1f41269534a..4476f432b98 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -507,6 +507,8 @@ int r600_bc_add_literal(struct r600_bc *bc, const u32 *value)
 	}
 	/* same on EG */
 	if (((bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3)) &&
+	     (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3)) &&
+	     (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3)) &&
 	     (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3))) ||
 		LIST_IS_EMPTY(&bc->cf_last->alu)) {
 		R600_ERR("last CF is not ALU (%p)\n", bc->cf_last);
@@ -728,6 +730,8 @@ static int r600_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 
 	switch (cf->inst) {
 	case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3):
+	case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
+	case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
 	case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
 		bc->bytecode[id++] = S_SQ_CF_ALU_WORD0_ADDR(cf->addr >> 1) |
 			S_SQ_CF_ALU_WORD0_KCACHE_MODE0(cf->kcache0_mode) |
@@ -808,6 +812,8 @@ int r600_bc_build(struct r600_bc *bc)
 	LIST_FOR_EACH_ENTRY(cf, &bc->cf, list) {
 		switch (cf->inst) {
 		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3):
+		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
+		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
 		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
 			break;
 		case V_SQ_CF_WORD1_SQ_CF_INST_TEX:
@@ -854,6 +860,8 @@ int r600_bc_build(struct r600_bc *bc)
 			return r;
 		switch (cf->inst) {
 		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3):
+		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
+		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
 		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
 			LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
 				switch(bc->chiprev) {
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 6cf23535515..aaebd16a310 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -2676,9 +2676,25 @@ static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode)
 
 static int pops(struct r600_shader_ctx *ctx, int pops)
 {
-	r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP));
-	ctx->bc->cf_last->pop_count = pops;
-	ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2;
+	int alu_pop = 3;
+	if (ctx->bc->cf_last) {
+		if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU) << 3)
+			alu_pop = 0;
+		else if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3)
+			alu_pop = 1;
+	}
+	alu_pop += pops;
+	if (alu_pop == 1) {
+		ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3;
+		ctx->bc->force_add_cf = 1;
+	} else if (alu_pop == 2) {
+		ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER) << 3;
+		ctx->bc->force_add_cf = 1;
+	} else {
+		r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP));
+		ctx->bc->cf_last->pop_count = pops;
+		ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2;
+	}
 	return 0;
 }