mesa: Bump version to 7.11-rc3

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
(cherry picked from commit 3daa2d97eb)

Makefile

@@ -183,7 +183,7 @@ ultrix-gcc:
 
 # Rules for making release tarballs
 
-VERSION=7.11-rc2
+VERSION=7.11-rc3
 DIRECTORY = Mesa-$(VERSION)
 LIB_NAME = MesaLib-$(VERSION)
 GLUT_NAME = MesaGLUT-$(VERSION)
@@ -481,13 +481,13 @@ rm_config: parsers
 	rm -f configs/autoconf
 
 $(LIB_NAME).tar: rm_config
-	cd .. ; tar -cf $(DIRECTORY)/$(LIB_NAME).tar $(LIB_FILES)
+	cd .. ; tar --dereference -cf $(DIRECTORY)/$(LIB_NAME).tar $(LIB_FILES)
 
 $(LIB_NAME).tar.gz: $(LIB_NAME).tar
 	gzip --stdout --best $(LIB_NAME).tar > $(LIB_NAME).tar.gz
 
 $(GLUT_NAME).tar:
-	cd .. ; tar -cf $(DIRECTORY)/$(GLUT_NAME).tar $(GLUT_FILES)
+	cd .. ; tar --dereference -cf $(DIRECTORY)/$(GLUT_NAME).tar $(GLUT_FILES)
 
 $(GLUT_NAME).tar.gz: $(GLUT_NAME).tar
 	gzip --stdout --best $(GLUT_NAME).tar > $(GLUT_NAME).tar.gz

src/glsl/lower_variable_index_to_cond_assign.cpp

@@ -29,6 +29,21 @@
  *
  * Pre-DX10 GPUs often don't have a native way to do this operation,
  * and this works around that.
+ *
+ * The lowering process proceeds as follows.  Each non-constant index
+ * found in an r-value is converted to a canonical form \c array[i].  Each
+ * element of the array is conditionally assigned to a temporary by comparing
+ * \c i to a constant index.  This is done by cloning the canonical form and
+ * replacing all occurances of \c i with a constant.  Each remaining occurance
+ * of the canonical form in the IR is replaced with a dereference of the
+ * temporary variable.
+ *
+ * L-values with non-constant indices are handled similarly.  In this case,
+ * the RHS of the assignment is assigned to a temporary.  The non-constant
+ * index is replace with the canonical form (just like for r-values).  The
+ * temporary is conditionally assigned to each element of the canonical form
+ * by comparing \c i with each index.  The same clone-and-replace scheme is
+ * used.
  */
 
 #include "ir.h"
@@ -37,10 +52,76 @@
 #include "glsl_types.h"
 #include "main/macros.h"
 
+static inline bool
+is_array_or_matrix(const ir_instruction *ir)
+{
+   return (ir->type->is_array() || ir->type->is_matrix());
+}
+
+/**
+ * Replace a dereference of a variable with a specified r-value
+ *
+ * Each time a dereference of the specified value is replaced, the r-value
+ * tree is cloned.
+ */
+class deref_replacer : public ir_rvalue_visitor {
+public:
+   deref_replacer(const ir_variable *variable_to_replace, ir_rvalue *value)
+      : variable_to_replace(variable_to_replace), value(value),
+	progress(false)
+   {
+      assert(this->variable_to_replace != NULL);
+      assert(this->value != NULL);
+   }
+
+   virtual void handle_rvalue(ir_rvalue **rvalue)
+   {
+      ir_dereference_variable *const dv = (*rvalue)->as_dereference_variable();
+
+      if ((dv != NULL) && (dv->var == this->variable_to_replace)) {
+	 this->progress = true;
+	 *rvalue = this->value->clone(ralloc_parent(*rvalue), NULL);
+      }
+   }
+
+   const ir_variable *variable_to_replace;
+   ir_rvalue *value;
+   bool progress;
+};
+
+/**
+ * Find a variable index dereference of an array in an rvalue tree
+ */
+class find_variable_index : public ir_hierarchical_visitor {
+public:
+   find_variable_index()
+      : deref(NULL)
+   {
+      /* empty */
+   }
+
+   virtual ir_visitor_status visit_enter(ir_dereference_array *ir)
+   {
+      if (is_array_or_matrix(ir->array)
+	  && (ir->array_index->as_constant() == NULL)) {
+	 this->deref = ir;
+	 return visit_stop;
+      }
+
+      return visit_continue;
+   }
+
+   /**
+    * First array dereference found in the tree that has a non-constant index.
+    */
+   ir_dereference_array *deref;
+};
+
 struct assignment_generator
 {
    ir_instruction* base_ir;
-   ir_rvalue* array;
+   ir_dereference *rvalue;
+   ir_variable *old_index;
    bool is_write;
    unsigned int write_mask;
    ir_variable* var;
@@ -55,18 +136,23 @@ struct assignment_generator
        * underlying variable.
        */
       void *mem_ctx = ralloc_parent(base_ir);
-      ir_dereference *element =
-	 new(mem_ctx) ir_dereference_array(this->array->clone(mem_ctx, NULL),
-					   new(mem_ctx) ir_constant(i));
-      ir_rvalue *variable = new(mem_ctx) ir_dereference_variable(this->var);
 
-      ir_assignment *assignment;
-      if (is_write) {
-	 assignment = new(mem_ctx) ir_assignment(element, variable, condition,
-						 write_mask);
-      } else {
-	 assignment = new(mem_ctx) ir_assignment(variable, element, condition);
-      }
+      /* Clone the old r-value in its entirety.  Then replace any occurances of
+       * the old variable index with the new constant index.
+       */
+      ir_dereference *element = this->rvalue->clone(mem_ctx, NULL);
+      ir_constant *const index = new(mem_ctx) ir_constant(i);
+      deref_replacer r(this->old_index, index);
+      element->accept(&r);
+      assert(r.progress);
+
+      /* Generate a conditional assignment to (or from) the constant indexed
+       * array dereference.
+       */
+      ir_rvalue *variable = new(mem_ctx) ir_dereference_variable(this->var);
+      ir_assignment *const assignment = (is_write)
+	 ? new(mem_ctx) ir_assignment(element, variable, condition, write_mask)
+	 : new(mem_ctx) ir_assignment(variable, element, condition);
 
       list->push_tail(assignment);
    }
@@ -233,21 +319,18 @@ public:
    bool lower_temps;
    bool lower_uniforms;
 
-   bool is_array_or_matrix(const ir_instruction *ir) const
+   bool storage_type_needs_lowering(ir_dereference_array *deref) const
    {
-      return (ir->type->is_array() || ir->type->is_matrix());
-   }
-
-   bool needs_lowering(ir_dereference_array *deref) const
-   {
-      if (deref == NULL || deref->array_index->as_constant()
-	  || !is_array_or_matrix(deref->array))
-	 return false;
-
-      if (deref->array->ir_type == ir_type_constant)
+      /* If a variable isn't eventually the target of this dereference, then
+       * it must be a constant or some sort of anonymous temporary storage.
+       *
+       * FINISHME: Is this correct?  Most drivers treat arrays of constants as
+       * FINISHME: uniforms.  It seems like this should do the same.
+       */
+      const ir_variable *const var = deref->array->variable_referenced();
+      if (var == NULL)
 	 return this->lower_temps;
 
-      const ir_variable *const var = deref->array->variable_referenced();
       switch (var->mode) {
       case ir_var_auto:
       case ir_var_temporary:
@@ -267,8 +350,18 @@ public:
       return false;
    }
 
+   bool needs_lowering(ir_dereference_array *deref) const
+   {
+      if (deref == NULL || deref->array_index->as_constant()
+	  || !is_array_or_matrix(deref->array))
+	 return false;
+
+      return this->storage_type_needs_lowering(deref);
+   }
+
    ir_variable *convert_dereference_array(ir_dereference_array *orig_deref,
-					  ir_assignment* orig_assign)
+					  ir_assignment* orig_assign,
+					  ir_dereference *orig_base)
    {
       assert(is_array_or_matrix(orig_deref->array));
 
@@ -314,9 +407,12 @@ public:
 	 new(mem_ctx) ir_assignment(lhs, orig_deref->array_index, NULL);
       base_ir->insert_before(assign);
 
+      orig_deref->array_index = lhs->clone(mem_ctx, NULL);
+
       assignment_generator ag;
-      ag.array = orig_deref->array;
+      ag.rvalue = orig_base;
       ag.base_ir = base_ir;
+      ag.old_index = index;
       ag.var = var;
       if (orig_assign) {
 	 ag.is_write = true;
@@ -327,21 +423,40 @@ public:
 
       switch_generator sg(ag, index, 4, 4);
 
-      exec_list list;
-      sg.generate(0, length, &list);
-      base_ir->insert_before(&list);
+      /* If the original assignment has a condition, respect that original
+       * condition!  This is acomplished by wrapping the new conditional
+       * assignments in an if-statement that uses the original condition.
+       */
+      if ((orig_assign != NULL) && (orig_assign->condition != NULL)) {
+	 /* No need to clone the condition because the IR that it hangs on is
+	  * going to be removed from the instruction sequence.
+	  */
+	 ir_if *if_stmt = new(mem_ctx) ir_if(orig_assign->condition);
+
+	 sg.generate(0, length, &if_stmt->then_instructions);
+	 base_ir->insert_before(if_stmt);
+      } else {
+	 exec_list list;
+
+	 sg.generate(0, length, &list);
+	 base_ir->insert_before(&list);
+      }
 
       return var;
    }
 
    virtual void handle_rvalue(ir_rvalue **pir)
    {
+      if (this->in_assignee)
+	 return;
+
       if (!*pir)
          return;
 
       ir_dereference_array* orig_deref = (*pir)->as_dereference_array();
       if (needs_lowering(orig_deref)) {
-         ir_variable* var = convert_dereference_array(orig_deref, 0);
+         ir_variable *var =
+	    convert_dereference_array(orig_deref, NULL, orig_deref);
          assert(var);
          *pir = new(ralloc_parent(base_ir)) ir_dereference_variable(var);
          this->progress = true;
@@ -353,10 +468,11 @@ public:
    {
       ir_rvalue_visitor::visit_leave(ir);
 
-      ir_dereference_array *orig_deref = ir->lhs->as_dereference_array();
+      find_variable_index f;
+      ir->lhs->accept(&f);
 
-      if (needs_lowering(orig_deref)) {
-         convert_dereference_array(orig_deref, ir);
+      if ((f.deref != NULL) && storage_type_needs_lowering(f.deref)) {
+         convert_dereference_array(f.deref, ir, ir->lhs);
          ir->remove();
          this->progress = true;
       }
@@ -377,7 +493,17 @@ lower_variable_index_to_cond_assign(exec_list *instructions,
 					   lower_temp,
 					   lower_uniform);
 
-   visit_list_elements(&v, instructions);
+   /* Continue lowering until no progress is made.  If there are multiple
+    * levels of indirection (e.g., non-constant indexing of array elements and
+    * matrix columns of an array of matrix), each pass will only lower one
+    * level of indirection.
+    */
+   bool progress_ever = false;
+   do {
+      v.progress = false;
+      visit_list_elements(&v, instructions);
+      progress_ever = v.progress || progress_ever;
+   } while (v.progress);
 
-   return v.progress;
+   return progress_ever;
 }

src/glsl/lower_vec_index_to_cond_assign.cpp

@@ -171,21 +171,23 @@ ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 
    assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT);
 
+   exec_list list;
+
    /* Store the index to a temporary to avoid reusing its tree. */
    index = new(ir) ir_variable(glsl_type::int_type, "vec_index_tmp_i",
 			       ir_var_temporary);
-   ir->insert_before(index);
+   list.push_tail(index);
    deref = new(ir) ir_dereference_variable(index);
    assign = new(ir) ir_assignment(deref, orig_deref->array_index, NULL);
-   ir->insert_before(assign);
+   list.push_tail(assign);
 
    /* Store the RHS to a temporary to avoid reusing its tree. */
    var = new(ir) ir_variable(ir->rhs->type, "vec_index_tmp_v",
 			     ir_var_temporary);
-   ir->insert_before(var);
+   list.push_tail(var);
    deref = new(ir) ir_dereference_variable(var);
    assign = new(ir) ir_assignment(deref, ir->rhs, NULL);
-   ir->insert_before(assign);
+   list.push_tail(assign);
 
    /* Generate a conditional move of each vector element to the temp. */
    for (i = 0; i < orig_deref->array->type->vector_elements; i++) {
@@ -205,8 +207,25 @@ ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 
       deref = new(ir) ir_dereference_variable(var);
       assign = new(ir) ir_assignment(swizzle, deref, condition);
-      ir->insert_before(assign);
+      list.push_tail(assign);
    }
+
+   /* If the original assignment has a condition, respect that original
+    * condition!  This is acomplished by wrapping the new conditional
+    * assignments in an if-statement that uses the original condition.
+    */
+   if (ir->condition != NULL) {
+      /* No need to clone the condition because the IR that it hangs on is
+       * going to be removed from the instruction sequence.
+       */
+      ir_if *if_stmt = new(mem_ctx) ir_if(ir->condition);
+
+      list.move_nodes_to(&if_stmt->then_instructions);
+      ir->insert_before(if_stmt);
+   } else {
+      ir->insert_before(&list);
+   }
+
    ir->remove();
 
    this->progress = true;

src/mesa/drivers/dri/i965/brw_fs_emit.cpp

@@ -59,7 +59,8 @@ fs_visitor::generate_fb_write(fs_inst *inst)
 
 	 if (inst->target > 0) {
 	    /* Set the render target index for choosing BLEND_STATE. */
-	    brw_MOV(p, retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE, 0, 2),
+	    brw_MOV(p, retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE,
+					   inst->base_mrf, 2),
 			      BRW_REGISTER_TYPE_UD),
 		    brw_imm_ud(inst->target));
 	 }

src/mesa/drivers/dri/i965/brw_fs_visitor.cpp

@@ -595,9 +595,11 @@ fs_visitor::emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
       /* gen4's SIMD8 sampler always has the slots for u,v,r present. */
       mlen += 3;
    } else if (ir->op == ir_txd) {
+      this->result = reg_undef;
       ir->lod_info.grad.dPdx->accept(this);
       fs_reg dPdx = this->result;
 
+      this->result = reg_undef;
       ir->lod_info.grad.dPdy->accept(this);
       fs_reg dPdy = this->result;
 
@@ -778,9 +780,11 @@ fs_visitor::emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
       inst = emit(FS_OPCODE_TXL, dst);
       break;
    case ir_txd: {
+      this->result = reg_undef;
       ir->lod_info.grad.dPdx->accept(this);
       fs_reg dPdx = this->result;
 
+      this->result = reg_undef;
       ir->lod_info.grad.dPdy->accept(this);
       fs_reg dPdy = this->result;
 
@@ -842,6 +846,7 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
    }
 
    if (ir->shadow_comparitor && ir->op != ir_txd) {
+      this->result = reg_undef;
       ir->shadow_comparitor->accept(this);
       emit(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result);
       mlen += reg_width;
@@ -852,11 +857,13 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
    case ir_tex:
       break;
    case ir_txb:
+      this->result = reg_undef;
       ir->lod_info.bias->accept(this);
       emit(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result);
       mlen += reg_width;
       break;
    case ir_txl:
+      this->result = reg_undef;
       ir->lod_info.lod->accept(this);
       emit(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result);
       mlen += reg_width;
@@ -865,9 +872,11 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
       if (c->dispatch_width == 16)
 	 fail("Gen7 does not support sample_d/sample_d_c in SIMD16 mode.");
 
+      this->result = reg_undef;
       ir->lod_info.grad.dPdx->accept(this);
       fs_reg dPdx = this->result;
 
+      this->result = reg_undef;
       ir->lod_info.grad.dPdy->accept(this);
       fs_reg dPdy = this->result;
 
@@ -1062,6 +1071,7 @@ fs_visitor::visit(ir_texture *ir)
       if (hw_compare_supported) {
 	 inst->shadow_compare = true;
       } else {
+	 this->result = reg_undef;
 	 ir->shadow_comparitor->accept(this);
 	 fs_reg ref = this->result;
 

src/mesa/drivers/dri/i965/brw_vs_emit.c

@@ -1821,6 +1821,9 @@ accumulator_contains(struct brw_vs_compile *c, struct brw_reg val)
    if (val.address_mode != BRW_ADDRESS_DIRECT)
       return GL_FALSE;
 
+   if (val.negate || val.abs)
+      return GL_FALSE;
+
    switch (prev_insn->header.opcode) {
    case BRW_OPCODE_MOV:
    case BRW_OPCODE_MAC:
@@ -1980,9 +1983,22 @@ void brw_vs_emit(struct brw_vs_compile *c )
 	      const struct prog_src_register *src = &inst->SrcReg[i];
 	      index = src->Index;
 	      file = src->File;	
-	      if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src)
-		  args[i] = c->output_regs[index].reg;
-	      else
+	      if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src) {
+		 /* Can't just make get_arg "do the right thing" here because
+		  * other callers of get_arg and get_src_reg don't expect any
+		  * special behavior for the c->output_regs[index].used_in_src
+		  * case.
+		  */
+		 args[i] = c->output_regs[index].reg;
+		 args[i].dw1.bits.swizzle =
+		    BRW_SWIZZLE4(GET_SWZ(src->Swizzle, 0),
+				 GET_SWZ(src->Swizzle, 1),
+				 GET_SWZ(src->Swizzle, 2),
+				 GET_SWZ(src->Swizzle, 3));
+
+		 /* Note this is ok for non-swizzle ARB_vp instructions */
+		 args[i].negate = src->Negate ? 1 : 0;
+	      } else
                   args[i] = get_arg(c, inst, i);
 	  }
 
@@ -1993,7 +2009,11 @@ void brw_vs_emit(struct brw_vs_compile *c )
       index = inst->DstReg.Index;
       file = inst->DstReg.File;
       if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src)
-	  dst = c->output_regs[index].reg;
+	 /* Can't just make get_dst "do the right thing" here because other
+	  * callers of get_dst don't expect any special behavior for the
+	  * c->output_regs[index].used_in_src case.
+	  */
+	 dst = brw_writemask(c->output_regs[index].reg, inst->DstReg.WriteMask);
       else
 	  dst = get_dst(c, inst->DstReg);
 

src/mesa/drivers/dri/i965/brw_vtbl.c

@@ -211,6 +211,7 @@ static void brw_new_batch( struct intel_context *intel )
    intel->batch.need_workaround_flush = true;
 
    brw->vb.nr_current_buffers = 0;
+   brw->ib.type = -1;
 
    /* Mark that the current program cache BO has been used by the GPU.
     * It will be reallocated if we need to put new programs in for the

src/mesa/drivers/dri/i965/brw_wm_emit.c

@@ -1094,9 +1094,16 @@ void emit_tex(struct brw_wm_compile *c,
    if (intel->gen < 5 && c->dispatch_width == 8)
       nr_texcoords = 3;
 
-   /* For shadow comparisons, we have to supply u,v,r. */
-   if (shadow)
-      nr_texcoords = 3;
+   if (shadow) {
+      if (intel->gen < 7) {
+	 /* For shadow comparisons, we have to supply u,v,r. */
+	 nr_texcoords = 3;
+      } else {
+	 /* On Ivybridge, the shadow comparitor comes first. Just load it. */
+	 brw_MOV(p, brw_message_reg(cur_mrf), arg[2]);
+	 cur_mrf += mrf_per_channel;
+      }
+   }
 
    /* Emit the texcoords. */
    for (i = 0; i < nr_texcoords; i++) {
@@ -1113,7 +1120,7 @@ void emit_tex(struct brw_wm_compile *c,
    }
 
    /* Fill in the shadow comparison reference value. */
-   if (shadow) {
+   if (shadow && intel->gen < 7) {
       if (intel->gen >= 5) {
 	 /* Fill in the cube map array index value. */
 	 brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));

src/mesa/drivers/dri/intel/intel_batchbuffer.c

@@ -388,6 +388,7 @@ intel_batchbuffer_emit_mi_flush(struct intel_context *intel)
 	 OUT_BATCH(PIPE_CONTROL_INSTRUCTION_FLUSH |
 		   PIPE_CONTROL_WRITE_FLUSH |
 		   PIPE_CONTROL_DEPTH_CACHE_FLUSH |
+		   PIPE_CONTROL_TC_FLUSH |
 		   PIPE_CONTROL_NO_WRITE);
 	 OUT_BATCH(0); /* write address */
 	 OUT_BATCH(0); /* write data */

src/mesa/main/version.h

@@ -35,7 +35,7 @@ struct gl_context;
 #define MESA_MAJOR 7
 #define MESA_MINOR 11
 #define MESA_PATCH 0
-#define MESA_VERSION_STRING "7.11-rc2"
+#define MESA_VERSION_STRING "7.11-rc3"
 
 /* To make version comparison easy */
 #define MESA_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))

src/mesa/program/ir_to_mesa.cpp

@@ -134,7 +134,7 @@ src_reg::src_reg(dst_reg reg)
    this->index = reg.index;
    this->swizzle = SWIZZLE_XYZW;
    this->negate = 0;
-   this->reladdr = NULL;
+   this->reladdr = reg.reladdr;
 }
 
 dst_reg::dst_reg(src_reg reg)
@@ -1494,6 +1494,18 @@ ir_to_mesa_visitor::visit(ir_dereference_array *ir)
 	      this->result, src_reg_for_float(element_size));
       }
 
+      /* If there was already a relative address register involved, add the
+       * new and the old together to get the new offset.
+       */
+      if (src.reladdr != NULL)  {
+	 src_reg accum_reg = get_temp(glsl_type::float_type);
+
+	 emit(ir, OPCODE_ADD, dst_reg(accum_reg),
+	      index_reg, *src.reladdr);
+
+	 index_reg = accum_reg;
+      }
+
       src.reladdr = ralloc(mem_ctx, src_reg);
       memcpy(src.reladdr, &index_reg, sizeof(index_reg));
    }

src/mesa/program/prog_optimize.c

@@ -1319,6 +1319,15 @@ _mesa_simplify_cmp(struct gl_program * program)
 
          inst->Opcode = OPCODE_MOV;
          inst->SrcReg[0] = inst->SrcReg[1];
+
+	 /* Unused operands are expected to have the file set to
+	  * PROGRAM_UNDEFINED.  This is how _mesa_init_instructions initializes
+	  * all of the sources.
+	  */
+	 inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+	 inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+	 inst->SrcReg[2].File = PROGRAM_UNDEFINED;
+	 inst->SrcReg[2].Swizzle = SWIZZLE_NOOP;
       }
    }
    if (dbg) {