diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 0d17f75da7e..93dc1420111 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -696,6 +696,7 @@ static int r600_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 
 		bc->bytecode[id++] = S_SQ_CF_ALU_WORD1_CF_INST(cf->inst >> 3) |
 					S_SQ_CF_ALU_WORD1_BARRIER(1) |
+					S_SQ_CF_ALU_WORD1_USES_WATERFALL(cf->r6xx_uses_waterfall) |
 					S_SQ_CF_ALU_WORD1_COUNT((cf->ndw / 2) - 1);
 		break;
 	case V_SQ_CF_WORD1_SQ_CF_INST_TEX:
diff --git a/src/gallium/drivers/r600/r600_asm.h b/src/gallium/drivers/r600/r600_asm.h
index 62a46cb0e1e..6aadf72957b 100644
--- a/src/gallium/drivers/r600/r600_asm.h
+++ b/src/gallium/drivers/r600/r600_asm.h
@@ -127,6 +127,7 @@ struct r600_bc_cf {
 	unsigned			pop_count;
 	unsigned			cf_addr; /* control flow addr */
 	unsigned			kcache0_mode;
+	unsigned			r6xx_uses_waterfall;
 	struct list_head		alu;
 	struct list_head		tex;
 	struct list_head		vtx;
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 4da6850b0a9..eac46a7ae60 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -2313,6 +2313,7 @@ static int tgsi_arl(struct r600_shader_ctx *ctx)
 	r = r600_bc_add_alu_type(ctx->bc, &alu, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU));
 	if (r)
 		return r;
+	ctx->bc->cf_last->r6xx_uses_waterfall = 1;
 	return 0;
 }