i965: Don't make consumers of brw_CONT/brw_WHILE track if depth in loop.

The codegen backends all had this same tracking, so just do it at the
EU level.

Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>

6 files changed, 25 insertions, 58 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_eu.c b/src/mesa/drivers/dri/i965/brw_eu.c
index c0126ff9ff..83aae3ba4d 100644
--- a/src/mesa/drivers/dri/i965/brw_eu.c
+++ b/src/mesa/drivers/dri/i965/brw_eu.c
@@ -197,6 +197,7 @@ brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
    p->loop_stack_depth = 0;
    p->loop_stack_array_size = 16;
    p->loop_stack = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
+   p->if_depth_in_loop = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h
index 8d06fefb57..11e7161dc6 100644
--- a/src/mesa/drivers/dri/i965/brw_eu.h
+++ b/src/mesa/drivers/dri/i965/brw_eu.h
@@ -134,6 +134,12 @@ struct brw_compile {
     * encountered.
     */
    int *loop_stack;
+   /**
+    * pre-gen6, the BREAK and CONT instructions had to tell how many IF/ENDIF
+    * blocks they were popping out of, to fix up the mask stack.  This tracks
+    * the IF/ENDIF nesting in each current nested loop level.
+    */
+   int *if_depth_in_loop;
    int loop_stack_depth;
    int loop_stack_array_size;
 
@@ -1026,8 +1032,8 @@ struct brw_instruction *brw_DO(struct brw_compile *p,
 
 struct brw_instruction *brw_WHILE(struct brw_compile *p);
 
-struct brw_instruction *brw_BREAK(struct brw_compile *p, int pop_count);
-struct brw_instruction *brw_CONT(struct brw_compile *p, int pop_count);
+struct brw_instruction *brw_BREAK(struct brw_compile *p);
+struct brw_instruction *brw_CONT(struct brw_compile *p);
 struct brw_instruction *gen6_CONT(struct brw_compile *p);
 /* Forward jumps:
  */
diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index 30ae3bb2ac..11f40807db 100644
--- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
@@ -918,10 +918,13 @@ push_loop_stack(struct brw_compile *p, struct brw_instruction *inst)
       p->loop_stack_array_size *= 2;
       p->loop_stack = reralloc(p->mem_ctx, p->loop_stack, int,
 			       p->loop_stack_array_size);
+      p->if_depth_in_loop = reralloc(p->mem_ctx, p->if_depth_in_loop, int,
+				     p->loop_stack_array_size);
    }
 
    p->loop_stack[p->loop_stack_depth] = inst - p->store;
    p->loop_stack_depth++;
+   p->if_depth_in_loop[p->loop_stack_depth] = 0;
 }
 
 static struct brw_instruction *
@@ -980,6 +983,7 @@ brw_IF(struct brw_compile *p, GLuint execute_size)
    p->current->header.predicate_control = BRW_PREDICATE_NONE;
 
    push_if_stack(p, insn);
+   p->if_depth_in_loop[p->loop_stack_depth]++;
    return insn;
 }
 
@@ -1187,6 +1191,7 @@ brw_ENDIF(struct brw_compile *p)
    struct brw_instruction *if_inst = NULL;
 
    /* Pop the IF and (optional) ELSE instructions from the stack */
+   p->if_depth_in_loop[p->loop_stack_depth]--;
    p->if_stack_depth--;
    if (p->if_stack[p->if_stack_depth]->header.opcode == BRW_OPCODE_ELSE) {
       else_inst = p->if_stack[p->if_stack_depth];
@@ -1245,7 +1250,7 @@ brw_ENDIF(struct brw_compile *p)
    patch_IF_ELSE(p, if_inst, else_inst, insn);
 }
 
-struct brw_instruction *brw_BREAK(struct brw_compile *p, int pop_count)
+struct brw_instruction *brw_BREAK(struct brw_compile *p)
 {
    struct intel_context *intel = &p->brw->intel;
    struct brw_instruction *insn;
@@ -1260,7 +1265,7 @@ struct brw_instruction *brw_BREAK(struct brw_compile *p, int pop_count)
       brw_set_src0(p, insn, brw_ip_reg());
       brw_set_src1(p, insn, brw_imm_d(0x0));
       insn->bits3.if_else.pad0 = 0;
-      insn->bits3.if_else.pop_count = pop_count;
+      insn->bits3.if_else.pop_count = p->if_depth_in_loop[p->loop_stack_depth];
    }
    insn->header.compression_control = BRW_COMPRESSION_NONE;
    insn->header.execution_size = BRW_EXECUTE_8;
@@ -1284,7 +1289,7 @@ struct brw_instruction *gen6_CONT(struct brw_compile *p)
    return insn;
 }
 
-struct brw_instruction *brw_CONT(struct brw_compile *p, int pop_count)
+struct brw_instruction *brw_CONT(struct brw_compile *p)
 {
    struct brw_instruction *insn;
    insn = next_insn(p, BRW_OPCODE_CONTINUE);
@@ -1295,7 +1300,7 @@ struct brw_instruction *brw_CONT(struct brw_compile *p, int pop_count)
    insn->header.execution_size = BRW_EXECUTE_8;
    /* insn->header.mask_control = BRW_MASK_DISABLE; */
    insn->bits3.if_else.pad0 = 0;
-   insn->bits3.if_else.pop_count = pop_count;
+   insn->bits3.if_else.pop_count = p->if_depth_in_loop[p->loop_stack_depth];
    return insn;
 }
 
diff --git a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
index 91e6961982..b68d8cb867 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
@@ -656,12 +656,6 @@ fs_visitor::generate_code()
    const char *last_annotation_string = NULL;
    ir_instruction *last_annotation_ir = NULL;
 
-   int loop_stack_array_size = 16;
-   int loop_stack_depth = 0;
-   int *if_depth_in_loop =
-      rzalloc_array(this->mem_ctx, int, loop_stack_array_size);
-
-
    if (unlikely(INTEL_DEBUG & DEBUG_WM)) {
       printf("Native code for fragment shader %d (%d-wide dispatch):\n",
 	     prog->Name, c->dispatch_width);
@@ -781,7 +775,6 @@ fs_visitor::generate_code()
 	 } else {
 	    brw_IF(p, c->dispatch_width == 16 ? BRW_EXECUTE_16 : BRW_EXECUTE_8);
 	 }
-	 if_depth_in_loop[loop_stack_depth]++;
 	 break;
 
       case BRW_OPCODE_ELSE:
@@ -789,22 +782,14 @@ fs_visitor::generate_code()
 	 break;
       case BRW_OPCODE_ENDIF:
 	 brw_ENDIF(p);
-	 if_depth_in_loop[loop_stack_depth]--;
 	 break;
 
       case BRW_OPCODE_DO:
 	 brw_DO(p, BRW_EXECUTE_8);
-	 loop_stack_depth++;
-	 if (loop_stack_array_size <= loop_stack_depth) {
-	    loop_stack_array_size *= 2;
-	    if_depth_in_loop = reralloc(this->mem_ctx, if_depth_in_loop, int,
-				        loop_stack_array_size);
-	 }
-	 if_depth_in_loop[loop_stack_depth] = 0;
 	 break;
 
       case BRW_OPCODE_BREAK:
-	 brw_BREAK(p, if_depth_in_loop[loop_stack_depth]);
+	 brw_BREAK(p);
 	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 	 break;
       case BRW_OPCODE_CONTINUE:
@@ -812,13 +797,11 @@ fs_visitor::generate_code()
 	 if (intel->gen >= 6)
 	    gen6_CONT(p);
 	 else
-	    brw_CONT(p, if_depth_in_loop[loop_stack_depth]);
+	    brw_CONT(p);
 	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 	 break;
 
       case BRW_OPCODE_WHILE:
-	 assert(loop_stack_depth > 0);
-	 loop_stack_depth--;
 	 brw_WHILE(p);
 	 break;
 
@@ -923,8 +906,6 @@ fs_visitor::generate_code()
       printf("\n");
    }
 
-   ralloc_free(if_depth_in_loop);
-
    brw_set_uip_jip(p);
 
    /* OK, while the INTEL_DEBUG=wm above is very nice for debugging FS
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp b/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
index 8ea5c14075..a618614c73 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
@@ -798,12 +798,6 @@ vec4_visitor::generate_code()
    const char *last_annotation_string = NULL;
    ir_instruction *last_annotation_ir = NULL;
 
-   int loop_stack_array_size = 16;
-   int loop_stack_depth = 0;
-   int *if_depth_in_loop =
-      rzalloc_array(this->mem_ctx, int, loop_stack_array_size);
-
-
    if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
       printf("Native code for vertex shader %d:\n", prog->Name);
    }
@@ -917,7 +911,6 @@ vec4_visitor::generate_code()
 	    struct brw_instruction *brw_inst = brw_IF(p, BRW_EXECUTE_8);
 	    brw_inst->header.predicate_control = inst->predicate;
 	 }
-	 if_depth_in_loop[loop_stack_depth]++;
 	 break;
 
       case BRW_OPCODE_ELSE:
@@ -925,22 +918,14 @@ vec4_visitor::generate_code()
 	 break;
       case BRW_OPCODE_ENDIF:
 	 brw_ENDIF(p);
-	 if_depth_in_loop[loop_stack_depth]--;
 	 break;
 
       case BRW_OPCODE_DO:
 	 brw_DO(p, BRW_EXECUTE_8);
-	 loop_stack_depth++;
-	 if (loop_stack_array_size <= loop_stack_depth) {
-	    loop_stack_array_size *= 2;
-	    if_depth_in_loop = reralloc(this->mem_ctx, if_depth_in_loop, int,
-				        loop_stack_array_size);
-	 }
-	 if_depth_in_loop[loop_stack_depth] = 0;
 	 break;
 
       case BRW_OPCODE_BREAK:
-	 brw_BREAK(p, if_depth_in_loop[loop_stack_depth]);
+	 brw_BREAK(p);
 	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 	 break;
       case BRW_OPCODE_CONTINUE:
@@ -948,13 +933,11 @@ vec4_visitor::generate_code()
 	 if (intel->gen >= 6)
 	    gen6_CONT(p);
 	 else
-	    brw_CONT(p, if_depth_in_loop[loop_stack_depth]);
+	    brw_CONT(p);
 	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 	 break;
 
       case BRW_OPCODE_WHILE:
-	 assert(loop_stack_depth > 0);
-	 loop_stack_depth--;
 	 brw_WHILE(p);
 	 break;
 
@@ -983,8 +966,6 @@ vec4_visitor::generate_code()
       printf("\n");
    }
 
-   ralloc_free(if_depth_in_loop);
-
    brw_set_uip_jip(p);
 
    /* OK, while the INTEL_DEBUG=vs above is very nice for debugging VS
diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c
index 62adc54c52..2b4b13a10d 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c
@@ -1843,8 +1843,7 @@ void brw_old_vs_emit(struct brw_vs_compile *c )
    struct brw_context *brw = p->brw;
    struct intel_context *intel = &brw->intel;
    const GLuint nr_insns = c->vp->program.Base.NumInstructions;
-   GLuint insn, loop_depth = 0;
-   int if_depth_in_loop[MAX_LOOP_DEPTH];
+   GLuint insn;
    const struct brw_indirect stack_index = brw_indirect(0, 0);   
    GLuint index;
    GLuint file;
@@ -1858,7 +1857,6 @@ void brw_old_vs_emit(struct brw_vs_compile *c )
 
    brw_set_compression_control(p, BRW_COMPRESSION_NONE);
    brw_set_access_mode(p, BRW_ALIGN_16);
-   if_depth_in_loop[loop_depth] = 0;
 
    brw_set_acc_write_control(p, 1);
 
@@ -2080,7 +2078,6 @@ void brw_old_vs_emit(struct brw_vs_compile *c )
 	 struct brw_instruction *if_inst = brw_IF(p, BRW_EXECUTE_8);
 	 /* Note that brw_IF smashes the predicate_control field. */
 	 if_inst->header.predicate_control = get_predicate(inst);
-	 if_depth_in_loop[loop_depth]++;
 	 break;
       }
       case OPCODE_ELSE:
@@ -2090,17 +2087,14 @@ void brw_old_vs_emit(struct brw_vs_compile *c )
       case OPCODE_ENDIF:
 	 clear_current_const(c);
 	 brw_ENDIF(p);
-	 if_depth_in_loop[loop_depth]--;
 	 break;			
       case OPCODE_BGNLOOP:
 	 clear_current_const(c);
 	 brw_DO(p, BRW_EXECUTE_8);
-         loop_depth++;
-	 if_depth_in_loop[loop_depth] = 0;
          break;
       case OPCODE_BRK:
 	 brw_set_predicate_control(p, get_predicate(inst));
-	 brw_BREAK(p, if_depth_in_loop[loop_depth]);
+	 brw_BREAK(p);
 	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
          break;
       case OPCODE_CONT:
@@ -2108,14 +2102,13 @@ void brw_old_vs_emit(struct brw_vs_compile *c )
 	 if (intel->gen >= 6) {
 	    gen6_CONT(p);
 	 } else {
-	    brw_CONT(p, if_depth_in_loop[loop_depth]);
+	    brw_CONT(p);
 	 }
          brw_set_predicate_control(p, BRW_PREDICATE_NONE);
          break;
 
       case OPCODE_ENDLOOP:
 	 clear_current_const(c);
-	 loop_depth--;
 	 brw_WHILE(p);
          break;