[Mesa-dev] [PATCH 03/12] nir: Add a loop analysis pass

Timothy Arceri timothy.arceri at collabora.com
Sun Aug 28 04:25:55 UTC 2016


On Sat, 2016-08-27 at 14:08 +0200, Thomas Helland wrote:
> 2016-08-27 8:03 GMT+02:00 Timothy Arceri <timothy.arceri at collabora.co
> m>:
> > 
> > This pass detects induction variables and calculates the
> > trip count of loops to be used for loop unrolling.
> > 
> > I've removed support for float induction values for now, for the
> > simple reason that they don't appear in my shader-db collection,
> > and so I don't see it as common enough that we want to pollute the
> > pass with this in the initial version.
> > 
> > V2: Rebase, adapt to removal of function overloads
> > 
> > V3: (Timothy Arceri)
> >  - don't try to find trip count if loop terminator conditional is a
> > phi
> >  - fix trip count for do-while loops
> >  - replace conditional type != alu assert with return
> >  - disable unrolling of loops with continues
> >  - multiple fixes to memory allocation, stop leaking and don't
> > destroy
> >    structs we want to use for unrolling.
> >  - find induction var when copy propagation disabled
> >  - fix iteration count bugs when induction var not on RHS of
> > condition
> >  - add FIXME for && conditions
> >  - calculate trip count for unsigned induction/limit vars
> > ---
> >  src/compiler/Makefile.sources       |   2 +
> >  src/compiler/nir/nir.h              |  75 +++
> >  src/compiler/nir/nir_loop_analyze.c | 922
> > ++++++++++++++++++++++++++++++++++++
> >  3 files changed, 999 insertions(+)
> >  create mode 100644 src/compiler/nir/nir_loop_analyze.c
> > 
> > diff --git a/src/compiler/Makefile.sources
> > b/src/compiler/Makefile.sources
> > index cfb6359..1bf1c52 100644
> > --- a/src/compiler/Makefile.sources
> > +++ b/src/compiler/Makefile.sources
> > @@ -187,6 +187,8 @@ NIR_FILES = \
> >         nir/nir_intrinsics.c \
> >         nir/nir_intrinsics.h \
> >         nir/nir_liveness.c \
> > +       nir/nir_loop_analyze.c \
> > +       nir/nir_loop_analyze.h \
> >         nir/nir_lower_alu_to_scalar.c \
> >         nir/nir_lower_atomics.c \
> >         nir/nir_lower_bitmap.c \
> > diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> > index e3e06b1..f85e829 100644
> > --- a/src/compiler/nir/nir.h
> > +++ b/src/compiler/nir/nir.h
> > @@ -1541,10 +1541,78 @@ nir_if_last_else_node(nir_if *if_stmt)
> >     return exec_node_data(nir_cf_node, tail, node);
> >  }
> > 
> > +typedef enum {
> > +   undefined,
> > +   invariant,
> > +   basic_induction
> > +} nir_loop_variable_type;
> > +
> > +typedef struct {
> > +   /* The ssa_def associated with this info */
> > +   nir_ssa_def *def;
> > +
> > +   /* The type of this ssa_def */
> > +   nir_loop_variable_type type;
> > +
> > +   /* Link to the loop_variable list for the loop */
> > +   struct list_head loop_vars_link;
> > +
> > +   /* A link for a list of invariant variables */
> > +   struct list_head invariant_link;
> > +
> > +   /* A link for a list of induction variables */
> > +   struct list_head induction_link;
> > +
> > +   /* If the ssa-def is constant */
> > +   bool is_constant;
> > +
> > +   bool in_conditional_block;
> > +
> > +   bool in_nested_loop;
> > +} nir_loop_variable;
> > +
> > +typedef struct {
> > +   nir_op alu_op;                                /* The type of
> > alu-operation    */
> > +   nir_loop_variable *alu_def;                   /* The def of the
> > alu-operation */
> > +   nir_loop_variable *invariant;                 /* The invariant
> > alu-operand    */
> > +   nir_loop_variable *phi;                       /* The other alu-
> > operand        */
> > +   nir_loop_variable *def_outside_loop;          /* The phi-src
> > outside the loop */
> > +} nir_basic_induction_var;
> > +
> > +typedef struct {
> > +   nir_if *nif;
> > +
> > +   /* Some more suitable fields like maybe indicated trip-count?
> > */
> > +   nir_instr *conditional_instr;
> 
> This question here should probably be answered, or removed.
> 
> > 
> > +
> > +   struct list_head loop_terminator_link;
> > +} nir_loop_terminator;
> > +
> > +typedef struct {
> > +   /* Loop_variable for all ssa_defs in loop */
> > +   struct list_head loop_vars_list;
> > +
> > +   /* How many times the loop is run (if known) */
> > +   uint32_t trip_count;
> > +   bool is_trip_count_known;
> > +
> > +   nir_loop_terminator *limiting_terminator;
> > +
> > +   /* A list of loop_terminators terminating this loop XXX: These
> > (apart from the limiting terminator) can be dead-code-eliminated */
> > +   struct list_head loop_terminator_list;
> > +
> 
> Line wrapping.
> 
> > 
> > +   /* The ssa_defs that are invariant */
> > +   struct list_head invariant_list;
> > +
> > +   struct hash_table *var_to_basic_ind;
> > +} nir_loop_info;
> > +
> >  typedef struct {
> >     nir_cf_node cf_node;
> > 
> >     struct exec_list body; /** < list of nir_cf_node */
> > +
> > +   nir_loop_info *info;
> >  } nir_loop;
> > 
> >  static inline nir_cf_node *
> > @@ -1569,6 +1637,7 @@ typedef enum {
> >     nir_metadata_dominance = 0x2,
> >     nir_metadata_live_ssa_defs = 0x4,
> >     nir_metadata_not_properly_reset = 0x8,
> > +   nir_metadata_loop_analysis = 0x16,
> >  } nir_metadata;
> > 
> >  typedef struct {
> > @@ -1751,6 +1820,8 @@ typedef struct nir_shader_compiler_options {
> >      * information must be inferred from the list of input
> > nir_variables.
> >      */
> >     bool use_interpolated_input_intrinsics;
> > +
> > +   unsigned max_unroll_iterations;
> >  } nir_shader_compiler_options;
> > 
> >  typedef struct nir_shader_info {
> > @@ -2542,6 +2613,10 @@ void nir_lower_double_pack(nir_shader
> > *shader);
> >  bool nir_normalize_cubemap_coords(nir_shader *shader);
> > 
> >  void nir_live_ssa_defs_impl(nir_function_impl *impl);
> > +
> > +void nir_loop_analyze_impl(nir_function_impl *impl);
> > +void nir_loop_analyze(nir_shader *shader);
> > +
> >  bool nir_ssa_defs_interfere(nir_ssa_def *a, nir_ssa_def *b);
> > 
> >  void nir_convert_to_ssa_impl(nir_function_impl *impl);
> > diff --git a/src/compiler/nir/nir_loop_analyze.c
> > b/src/compiler/nir/nir_loop_analyze.c
> > new file mode 100644
> > index 0000000..a01a4b5
> > --- /dev/null
> > +++ b/src/compiler/nir/nir_loop_analyze.c
> > @@ -0,0 +1,922 @@
> > +/*
> > + * Copyright © 2015 Thomas Helland (for Google's Summer of Code)
> > + *
> > + * Permission is hereby granted, free of charge, to any person
> > obtaining a
> > + * copy of this software and associated documentation files (the
> > "Software"),
> > + * to deal in the Software without restriction, including without
> > limitation
> > + * the rights to use, copy, modify, merge, publish, distribute,
> > sublicense,
> > + * and/or sell copies of the Software, and to permit persons to
> > whom the
> > + * Software is furnished to do so, subject to the following
> > conditions:
> > + *
> > + * The above copyright notice and this permission notice
> > (including the next
> > + * paragraph) shall be included in all copies or substantial
> > portions of the
> > + * Software.
> > + *
> > + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
> > EXPRESS OR
> > + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> > MERCHANTABILITY,
> > + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO
> > EVENT SHALL
> > + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
> > DAMAGES OR OTHER
> > + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> > ARISING
> > + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> > OTHER DEALINGS
> > + * IN THE SOFTWARE.
> > + */
> > +
> > +#include "nir.h"
> > +
> > +typedef struct {
> > +   /* A link for the work list */
> > +   struct list_head process_link;
> > +
> > +   bool in_loop;
> > +
> > +   nir_loop_variable *nir_loop_var;
> > +} loop_variable;
> > +
> > +typedef struct {
> > +   bool contains_break;
> > +   bool contains_continue;
> > +} loop_jumps;
> > +
> > +typedef struct {
> > +   /* The loop we store information for */
> > +   nir_loop *loop;
> > +
> > +   /* Loop_variable for all ssa_defs in function */
> > +   loop_variable *loop_vars;
> > +
> > +   /* Loop_variable for all ssa_defs in function */
> > +   nir_loop_variable *nir_loop_vars;
> > +
> > +   /* A list of the loop_vars to analyze */
> > +   struct list_head process_list;
> > +
> > +   nir_loop_info *info;
> > +} loop_info_state;
> > +
> > +static loop_variable *
> > +get_loop_var(nir_ssa_def *value, loop_info_state *state)
> > +{
> > +   return &(state->loop_vars[value->index]);
> > +}
> > +
> > +static nir_loop_variable *
> > +get_nir_loop_var(nir_ssa_def *value, loop_info_state *state)
> > +{
> > +   return &(state->nir_loop_vars[value->index]);
> > +}
> > +
> > +typedef struct {
> > +   loop_info_state *state;
> > +   bool mark_nested;
> > +   bool mark_in_conditional;
> > +} init_loop_state;
> > +
> > +static bool
> > +init_loop_def(nir_ssa_def *def, void *void_init_loop_state)
> > +{
> > +   init_loop_state *loop_init_state = void_init_loop_state;
> > +   loop_variable *var = get_loop_var(def, loop_init_state->state);
> > +
> > +   /* Add to the tail of the list. That way we start at the
> > beginning of the
> > +    * defs in the loop instead of the end when walking the list.
> > This means
> > +    * less recursive calls. Only add defs that are not in nested
> > loops or
> > +    * conditional blocks.
> > +    */
> > +   if (!(loop_init_state->mark_in_conditional ||
> > +         loop_init_state->mark_nested))
> > +      LIST_ADDTAIL(&(var->process_link),
> > +                   &(loop_init_state->state->process_list));
> > +
> > +   if (loop_init_state->mark_in_conditional)
> > +      var->nir_loop_var->in_conditional_block = true;
> > +
> > +   if (loop_init_state->mark_nested)
> > +      var->nir_loop_var->in_nested_loop = true;
> > +
> > +   var->in_loop = true;
> > +
> > +   return true;
> > +}
> > +
> > +static bool
> > +init_loop_block(nir_block *block, void *void_init_loop_state)
> > +{
> > +   init_loop_state *loop_init_state = void_init_loop_state;
> > +
> > +   nir_foreach_instr(instr, block)
> > +      nir_foreach_ssa_def(instr, init_loop_def, loop_init_state);
> > +
> > +   return true;
> > +}
> > +
> > +static inline bool
> > +is_var_alu(loop_variable *var)
> > +{
> > +   return (var->nir_loop_var->def->parent_instr->type ==
> > nir_instr_type_alu);
> > +}
> > +
> > +static inline bool
> > +is_var_phi(loop_variable *var)
> > +{
> > +   return (var->nir_loop_var->def->parent_instr->type ==
> > nir_instr_type_phi);
> > +}
> > +
> > +static inline bool
> > +is_ssa_def_invariant(nir_ssa_def *def, loop_info_state *state)
> > +{
> > +   loop_variable *var = get_loop_var(def, state);
> > +
> > +   if (var->nir_loop_var->type == invariant)
> > +      return true;
> > +
> > +   if (!var->in_loop) {
> > +      var->nir_loop_var->type = invariant;
> > +      return true;
> > +   }
> > +
> > +   if (var->nir_loop_var->type == basic_induction)
> > +      return false;
> > +
> > +   if (is_var_alu(var)) {
> > +      nir_alu_instr *alu = nir_instr_as_alu(def->parent_instr);
> > +
> > +      for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs;
> > i++) {
> > +         if (!is_ssa_def_invariant(alu->src[i].src.ssa, state))
> > +            return false;
> > +      }
> > +      var->nir_loop_var->type = invariant;
> > +      return true;
> > +   }
> > +
> > +   /* Phis shouldn't be invariant except if one operand is
> > invariant, and the
> > +    * other is the phi itself. These should be removed by
> > opt_remove_phis.
> > +    * load_consts are already set to invariant and constant in the
> > +    * initialization. Skip both of these, and the rest of the
> > operations.
> > +    */
> 
> This comment is kinda weird. A more correct wording would be:
> 
> /* Phis shouldn't be invariant except if one operand is invariant,
> and the
>  * other is the phi itself. These should be removed by
> opt_remove_phis.
>  * load_consts are already set to invariant and constant during init,
>  * and so should return earlier. Remaining op_codes are set
> undefined.
>  */
> 
> > 
> > +   var->nir_loop_var->type = undefined;
> > +   return false;
> > +}
> > +
> > +static void
> > +compute_invariance_information(loop_info_state *state)
> > +{
> > +   /* An expression is invariant in a loop L if:
> > +    *  (base cases)
> > +    *    – it’s a constant
> > +    *    – it’s a variable use, all of whose single defs are
> > outside of L
> > +    *  (inductive cases)
> > +    *    – it’s a pure computation all of whose args are loop
> > invariant
> > +    *    – it’s a variable use whose single reaching def, and the
> > +    *      rhs of that def is loop-invariant
> > +    */
> > +   bool changes;
> > +
> > +   do {
> > +      changes = false;
> > +      list_for_each_entry_safe(loop_variable, var,
> > +                               &state->process_list, process_link)
> > {
> > +
> > +         if (var->nir_loop_var->in_conditional_block ||
> > +             var->nir_loop_var->in_nested_loop) {
> > +            LIST_DEL(&var->process_link);
> > +            continue;
> > +         }
> > +
> > +         if (is_ssa_def_invariant(var->nir_loop_var->def, state))
> > {
> > +            LIST_DEL(&var->process_link);
> > +            changes = true;
> > +         }
> > +      }
> > +   } while (changes);
> > +}
> > +
> > +static inline bool
> > +is_var_basic_induction_var(loop_variable *var, loop_info_state
> > *state)
> > +{
> > +   if (var->nir_loop_var->type == basic_induction)
> > +      return true;
> > +
> > +   /* We are only interested in checking phi's for the basic
> > induction
> > +    * variable case as its simple to detect. All basic induction
> > variables
> > +    * have a phi node
> > +    */
> > +   if (!is_var_phi(var))
> > +      return false;
> > +
> > +   nir_phi_instr *phi = nir_instr_as_phi(var->nir_loop_var->def-
> > >parent_instr);
> > +
> > +   nir_basic_induction_var *biv = rzalloc(state,
> > nir_basic_induction_var);
> > +   biv->phi = var->nir_loop_var;
> > +
> > +   nir_foreach_phi_src(src, phi) {
> > +      loop_variable *src_var = get_loop_var(src->src.ssa, state);
> > +
> > +      /* If one of the sources is in a conditional or nested then
> > panic */
> 
> This comment is missing "block".
> 
> This was just a fast run through of the patch correcting some of the
> stuff I know I left behind. I'll try to get a more thorough look at
> the
> patch later on.


That would be great thanks. I'll fix up these and send out a new series
in a couple of days as I spotted a couple of things myself.

Thanks,
Tim

> 
> Regards,
> Thomas
> 
> > 
> > +      if (src_var->nir_loop_var->in_conditional_block ||
> > +          src_var->nir_loop_var->in_nested_loop)
> > +         break;
> > +
> > +      /* We disable copy propagation inside loops to simplify
> > unrolling so
> > +       * follow moves when looking for induction variable.
> > +       */
> > +      bool copy_prop;
> > +      nir_alu_instr *alu = NULL;
> > +      do {
> > +         copy_prop = false;
> > +         if (!src_var->in_loop)
> > +            biv->def_outside_loop = src_var->nir_loop_var;
> > +
> > +         if (src_var->in_loop && is_var_alu(src_var) &&
> > +             !src_var->nir_loop_var->in_nested_loop &&
> > +             !src_var->nir_loop_var->in_conditional_block) {
> > +
> > +            alu = nir_instr_as_alu(src_var->nir_loop_var->def-
> > >parent_instr);
> > +
> > +            if (alu->op == nir_op_imov || alu->op == nir_op_fmov)
> > {
> > +               src_var = get_loop_var(alu->src[0].src.ssa, state);
> > +               copy_prop = true;
> > +            }
> > +         }
> > +      } while (copy_prop);
> > +
> > +
> > +      if (src_var->in_loop && is_var_alu(src_var) &&
> > +          !src_var->nir_loop_var->in_nested_loop &&
> > +          !src_var->nir_loop_var->in_conditional_block) {
> > +         switch (alu->op) {
> > +         case nir_op_fadd:    case nir_op_iadd:    case
> > nir_op_uadd_carry:
> > +         case nir_op_fsub:    case nir_op_isub:    case
> > nir_op_usub_borrow:
> > +         case nir_op_fmul:    case nir_op_imul:    case
> > nir_op_umul_high:
> > +         case nir_op_fdiv:    case nir_op_idiv:    case
> > nir_op_udiv:
> > +
> > +            biv->alu_def = src_var->nir_loop_var;
> > +
> > +            for (unsigned i = 0; i < 2; i++) {
> > +               /* Is one of the operands invariant, and the other
> > the phi? */
> > +               if (is_ssa_def_invariant(alu->src[i].src.ssa,
> > state) &&
> > +                   alu->src[1-i].src.ssa->index == phi-
> > >dest.ssa.index)
> > +                  biv->invariant = get_nir_loop_var(alu-
> > >src[i].src.ssa,
> > +                                                    state);
> > +            }
> > +
> > +            biv->alu_op = alu->op;
> > +            break;
> > +
> > +         default:
> > +            break;
> > +         }
> > +      }
> > +   }
> > +
> > +   if (biv->alu_def && biv->def_outside_loop && biv->invariant &&
> > biv->phi) {
> > +      biv->alu_def->type = basic_induction;
> > +      biv->phi->type = basic_induction;
> > +      _mesa_hash_table_insert(state->info->var_to_basic_ind, biv-
> > >alu_def, biv);
> > +      _mesa_hash_table_insert(state->info->var_to_basic_ind, biv-
> > >phi, biv);
> > +      return true;
> > +   }
> > +
> > +   /* The requirements for a basic induction variable are not
> > fulfilled */
> > +   ralloc_free(biv);
> > +   return false;
> > +}
> > +
> > +static bool
> > +compute_induction_information(loop_info_state *state)
> > +{
> > +   bool changes;
> > +   bool found_induction_var = false;
> > +
> > +   do {
> > +      changes = false;
> > +      list_for_each_entry_safe(loop_variable, var,
> > +                               &state->process_list, process_link)
> > {
> > +
> > +         /* It can't be an induction variable if it is invariant.
> > We don't
> > +          * want to deal with things in nested loops or
> > conditionals.
> > +          */
> > +         if (var->nir_loop_var->type == invariant ||
> > +             var->nir_loop_var->in_conditional_block ||
> > +             var->nir_loop_var->in_nested_loop) {
> > +            LIST_DEL(&(var->process_link));
> > +            continue;
> > +         }
> > +
> > +         if (is_var_basic_induction_var(var, state)) {
> > +            /* If a phi is marked basic_ind we also mark the
> > alu_def basic_ind
> > +             * at the same time. It will then be detected as
> > basic_ind here,
> > +             * on the second passing, and be removed from the
> > list.
> > +             */
> > +            LIST_DEL(&(var->process_link));
> > +            changes = true;
> > +            found_induction_var = true;
> > +         }
> > +      }
> > +   } while (changes);
> > +
> > +   return found_induction_var;
> > +}
> > +
> > +static bool
> > +initialize_ssa_def(nir_ssa_def *def, void *void_state)
> > +{
> > +   loop_info_state *state = void_state;
> > +   loop_variable *var = get_loop_var(def, state);
> > +
> > +   var->nir_loop_var = get_nir_loop_var(def, state);
> > +
> > +   var->in_loop = false;
> > +   var->nir_loop_var->def = def;
> > +
> > +   if (def->parent_instr->type == nir_instr_type_load_const) {
> > +      var->nir_loop_var->type = invariant;
> > +      var->nir_loop_var->is_constant = true;
> > +   } else {
> > +      var->nir_loop_var->type = undefined;
> > +   }
> > +
> > +   return true;
> > +}
> > +
> > +static bool
> > +foreach_cf_node_ex_loop(nir_cf_node *node, void *state)
> > +{
> > +   nir_block *block;
> > +
> > +   switch (node->type) {
> > +   case nir_cf_node_block:
> > +      block = nir_cf_node_as_block(node);
> > +      nir_foreach_instr(instr, block) {
> > +         if (instr->type == nir_instr_type_jump) {
> > +            if (nir_instr_as_jump(instr)->type == nir_jump_break)
> > {
> > +               ((loop_jumps *) state)->contains_break = true;
> > +            } else if (nir_instr_as_jump(instr)->type ==
> > nir_jump_continue) {
> > +               ((loop_jumps *) state)->contains_continue = true;
> > +            }
> > +         }
> > +      }
> > +      return true;
> > +
> > +   case nir_cf_node_if: {
> > +      nir_if *if_stmt = nir_cf_node_as_if(node);
> > +
> > +      foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt-
> > >then_list)
> > +         if (!foreach_cf_node_ex_loop(node, state))
> > +            return false;
> > +
> > +      foreach_list_typed_safe(nir_cf_node, node, node, &if_stmt-
> > >else_list)
> > +         if (!foreach_cf_node_ex_loop(node, state))
> > +            return false;
> > +
> > +      break;
> > +   }
> > +
> > +   default:
> > +      break;
> > +   }
> > +
> > +   return false;
> > +}
> > +
> > +static bool
> > +is_trivial_loop_terminator(nir_if *nif)
> > +{
> > +   /* If there is stuff in the else-block that means that this is
> > not a
> > +    * simple break on true if-statement and so we bail
> > +    */
> > +   foreach_list_typed_safe(nir_cf_node, node, node, &nif-
> > >else_list)
> > +      if (node->type == nir_cf_node_block)
> > +         nir_foreach_instr(instr, nir_cf_node_as_block(node))
> > +            return false;
> > +
> > +   nir_cf_node *first_then = nir_if_first_then_node(nif);
> > +   nir_block *first_then_block = nir_cf_node_as_block(first_then);
> > +   nir_instr *first_instr =
> > nir_block_first_instr(first_then_block);
> > +
> > +   if (first_instr && first_instr->type == nir_instr_type_jump &&
> > +       nir_instr_as_jump(first_instr)->type == nir_jump_break) {
> > +      return true;
> > +   }
> > +
> > +   return false;
> > +}
> > +
> > +static bool
> > +find_loop_terminators(loop_info_state *state)
> > +{
> > +   bool success = false;
> > +   foreach_list_typed_safe(nir_cf_node, node, node, &state->loop-
> > >body) {
> > +      if (node->type == nir_cf_node_if) {
> > +         nir_if *nif = nir_cf_node_as_if(node);
> > +
> > +         /* Don't check the nested loops if there are breaks */
> > +         loop_jumps lj;
> > +         lj.contains_break = false;
> > +         lj.contains_continue = false;
> > +
> > +         foreach_cf_node_ex_loop(&nif->cf_node, &lj);
> > +
> > +         if (lj.contains_continue)
> > +            return false;
> > +
> > +         if (!lj.contains_break)
> > +            continue;
> > +
> > +         /* If there is a break then we should find a terminator.
> > If we can
> > +          * not find a loop terminator, but there is a break-
> > statement then
> > +          * we should return false so that we do not try to find
> > trip-count
> > +          */
> > +         if (!is_trivial_loop_terminator(nif))
> > +            return false;
> > +
> > +         if (nif->condition.ssa->parent_instr->type ==
> > nir_instr_type_phi)
> > +            return false;
> > +
> > +         nir_loop_terminator *terminator =
> > +            rzalloc(state->loop, nir_loop_terminator);
> > +
> > +         list_add(&terminator->loop_terminator_link,
> > +                  &state->info->loop_terminator_list);
> > +
> > +         terminator->nif = nif;
> > +         terminator->conditional_instr = nif->condition.ssa-
> > >parent_instr;
> > +
> > +         success = true;
> > +      }
> > +   }
> > +
> > +   return success;
> > +}
> > +
> > +static nir_basic_induction_var *
> > +get_basic_ind_var_for_loop_var(loop_variable *var, loop_info_state
> > *state)
> > +{
> > +   assert(var->nir_loop_var->type == basic_induction);
> > +
> > +   struct hash_entry *entry =
> > +         _mesa_hash_table_search(state->info->var_to_basic_ind,
> > +                                var->nir_loop_var);
> > +
> > +   return entry->data;
> > +}
> > +
> > +static int32_t
> > +get_iteration(nir_op cond_op, nir_const_value *initial,
> > nir_const_value *step,
> > +              nir_const_value *limit, nir_alu_instr *alu, int32_t
> > *init_val,
> > +              bool increment_before)
> > +{
> > +   int32_t iter;
> > +
> > +   switch (cond_op) {
> > +   case nir_op_ige:
> > +   case nir_op_ilt:
> > +   case nir_op_ieq:
> > +   case nir_op_ine: {
> > +      int32_t initial_val = initial->i32[0];
> > +      if (increment_before) {
> > +         initial_val = alu->op == nir_op_iadd ?
> > +            initial_val + step->i32[0] : initial_val - step-
> > >i32[0];
> > +      }
> > +
> > +      int32_t span = limit->i32[0] - initial_val;
> > +      iter = span / step->i32[0];
> > +      *init_val = initial_val;
> > +      break;
> > +   }
> > +   case nir_op_uge:
> > +   case nir_op_ult: {
> > +      uint32_t initial_val = initial->u32[0];
> > +      if (increment_before) {
> > +         initial_val = alu->op == nir_op_iadd ?
> > +            initial_val + step->u32[0] : initial_val - step-
> > >u32[0];
> > +      }
> > +
> > +      uint32_t span = limit->u32[0] - initial_val;
> > +      iter = span / step->u32[0];
> > +      *init_val = initial_val;
> > +      break;
> > +   }
> > +   default:
> > +      return -1;
> > +   }
> > +
> > +   return iter;
> > +}
> > +
> > +static uint32_t
> > +utest_interations(int32_t iter_int, nir_const_value *step,
> > +                  nir_const_value *limit, nir_op cond_op,
> > +                  uint32_t initial_val, bool limit_rhs)
> > +{
> > +   bool valid_loop = false;
> > +   uint32_t mul = iter_int * step->u32[0];
> > +   uint32_t uadd = mul + initial_val;
> > +
> > +   switch (cond_op) {
> > +   case nir_op_uge:
> > +      valid_loop = limit_rhs ? uadd >= limit->u32[0] : uadd <=
> > limit->u32[0];
> > +      break;
> > +   case nir_op_ult:
> > +      valid_loop = limit_rhs ? uadd < limit->u32[0] : uadd >
> > limit->u32[0];
> > +      break;
> > +   default:
> > +      unreachable("Unhandled loop condition!");
> > +   }
> > +
> > +   return valid_loop;
> > +}
> > +
> > +static int32_t
> > +itest_interations(int32_t iter_int, nir_const_value *step,
> > +                  nir_const_value *limit, nir_op cond_op,
> > +                  int32_t initial_val, bool limit_rhs)
> > +{
> > +   bool valid_loop = false;
> > +   int32_t mul = iter_int * step->i32[0];
> > +   int32_t iadd = mul + initial_val;
> > +
> > +   switch (cond_op) {
> > +   case nir_op_ige:
> > +      valid_loop = limit_rhs ? iadd >= limit->i32[0] : iadd <=
> > limit->i32[0];
> > +      break;
> > +   case nir_op_ilt:
> > +      valid_loop = limit_rhs ? iadd < limit->i32[0] : iadd >
> > limit->i32[0];
> > +      break;
> > +   case nir_op_ieq:
> > +      valid_loop = iadd == limit->i32[0];
> > +      break;
> > +   case nir_op_ine:
> > +      valid_loop = iadd != limit->i32[0];
> > +      break;
> > +   default:
> > +      unreachable("Unhandled loop condition!");
> > +   }
> > +
> > +   return valid_loop;
> > +}
> > +
> > +static int
> > +calculate_iterations(nir_const_value *initial, nir_const_value
> > *step,
> > +                     nir_const_value *limit, nir_op cond_op,
> > +                     nir_loop_variable *alu_def, nir_alu_instr
> > *cond_alu,
> > +                     bool limit_rhs)
> > +{
> > +   /* Superfluous assert */
> > +   assert(initial != NULL && step != NULL && limit != NULL);
> > +
> > +   nir_alu_instr *alu = nir_instr_as_alu(alu_def->def-
> > >parent_instr);
> > +
> > +   /* Unsupported alu operation */
> > +   if (!(alu->op == nir_op_iadd || alu->op == nir_op_isub))
> > +      return -1;
> > +
> > +   /* do-while loops can increment the starting value before the
> > condition is
> > +    * checked. e.g.
> > +    *
> > +    *    do {
> > +    *        ndx++;
> > +    *     } while (ndx < 3);
> > +    *
> > +    * Here we check if the induction variable is used directly by
> > the loop
> > +    * condition and if so we assume we need to step the initial
> > value.
> > +    */
> > +   bool increment_before = false;
> > +   if (cond_alu->src[0].src.ssa == alu_def->def || cond_alu-
> > >src[1].src.ssa == alu_def->def)
> > +      increment_before = true;
> > +
> > +   int32_t initial_val;
> > +   int iter_int = get_iteration(cond_op, initial, step, limit,
> > alu,
> > +                                &initial_val, increment_before);
> > +
> > +   /* If iter_int is negative the loop is ill-formed or is the
> > conditional is
> > +    * unsigned with a huge iteration count so don't bother going
> > any further.
> > +    */
> > +   if (iter_int < 0)
> > +      return -1;
> > +
> > +   /* An explanation from the GLSL unrolling pass:
> > +    *
> > +    * Make sure that the calculated number of iterations satisfies
> > the exit
> > +    * condition.  This is needed to catch off-by-one errors and
> > some types of
> > +    * ill-formed loops.  For example, we need to detect that the
> > following
> > +    * loop does not have a maximum iteration count.
> > +    *
> > +    *    for (float x = 0.0; x != 0.9; x += 0.2);
> > +    */
> > +   const int bias[] = { -1, 1, 1 };
> > +
> > +   for (unsigned i = 0; i < ARRAY_SIZE(bias); i++) {
> > +      iter_int = iter_int + bias[i];
> > +
> > +      switch (cond_op) {
> > +      case nir_op_ige:
> > +      case nir_op_ilt:
> > +      case nir_op_ieq:
> > +      case nir_op_ine:
> > +         if (itest_interations(iter_int, step, limit, cond_op,
> > initial_val,
> > +                               limit_rhs)) {
> > +            return iter_int;
> > +         }
> > +         break;
> > +      case nir_op_uge:
> > +      case nir_op_ult:
> > +         if (utest_interations(iter_int, step, limit, cond_op,
> > +                               (uint32_t) initial_val, limit_rhs))
> > {
> > +            return iter_int;
> > +         }
> > +         break;
> > +      default:
> > +         return -1;
> > +      }
> > +   }
> > +
> > +   return -1;
> > +}
> > +
> > +/* Run through each of the terminators of the loop and try to
> > infer a possible
> > + * trip-count. We need to check them all, and set the lowest trip-
> > count as the
> > + * trip-count of our loop. If one of the terminators has an
> > undecidable
> > + * trip-count we can not safely assume anything about the duration
> > of the
> > + * loop.
> > + */
> > +static void
> > +find_trip_count(loop_info_state *state)
> > +{
> > +   state->info->is_trip_count_known = false;
> > +   nir_loop_terminator *limiting_terminator = NULL;
> > +   int min_trip_count = -2;
> > +
> > +   list_for_each_entry(nir_loop_terminator, terminator,
> > +                       &state->info->loop_terminator_list,
> > loop_terminator_link) {
> > +
> > +      if (terminator->conditional_instr->type !=
> > nir_instr_type_alu) {
> > +         /* If we get here the loop is likely not really a loop
> > and will get
> > +          * cleaned up elsewhere.
> > +          */
> > +         return;
> > +      }
> > +
> > +      nir_alu_instr *alu = nir_instr_as_alu(terminator-
> > >conditional_instr);
> > +      loop_variable *basic_ind = NULL;
> > +      loop_variable *limit = NULL;
> > +      bool limit_rhs = true;
> > +      nir_op cond_op;
> > +
> > +      /* FIXME: We should also be able to handle nir_op_iand for
> > example
> > +       *   if ((i >= minLights) && (i >= lightProps_count)) break;
> > +       */
> > +      switch (alu->op) {
> > +      case nir_op_fge:      case nir_op_ige:      case nir_op_uge:
> > +      case nir_op_flt:      case nir_op_ilt:      case nir_op_ult:
> > +      case nir_op_feq:      case nir_op_ieq:
> > +      case nir_op_fne:      case nir_op_ine:
> > +
> > +         /* We assume that the limit is the "right" operand */
> > +         basic_ind = get_loop_var(alu->src[0].src.ssa, state);
> > +         limit = get_loop_var(alu->src[1].src.ssa, state);
> > +         cond_op = alu->op;
> > +
> > +         if (basic_ind->nir_loop_var->type != basic_induction) {
> > +            /* We had it the wrong way, flip things around */
> > +            basic_ind = get_loop_var(alu->src[1].src.ssa, state);
> > +            limit = get_loop_var(alu->src[0].src.ssa, state);
> > +            limit_rhs = false;
> > +         }
> > +
> > +         /* The comparison has to have a basic induction variable
> > +          * and a constant for us to be able to find trip counts
> > +          */
> > +         if (basic_ind->nir_loop_var->type != basic_induction ||
> > +             !limit->nir_loop_var->is_constant)
> > +            return;
> > +
> > +         nir_basic_induction_var *ind =
> > +               get_basic_ind_var_for_loop_var(basic_ind, state);
> > +
> > +         if (!ind->def_outside_loop->is_constant ||
> > +             !ind->invariant->is_constant)
> > +            return;
> > +
> > +         /* We have determined that we have the following
> > constants:
> > +          * (With the typical int i = 0; i < x; i++; as an
> > example)
> > +          *    - Upper limit.
> > +          *    - Starting value
> > +          *    - Step / iteration size
> > +          * Thats all thats needed to calculate the trip-count
> > +          */
> > +
> > +         nir_load_const_instr *initial_instr =
> > +               nir_instr_as_load_const(
> > +                     ind->def_outside_loop->def->parent_instr);
> > +
> > +         nir_const_value initial_val = initial_instr->value;
> > +
> > +         nir_load_const_instr *step_instr =
> > +               nir_instr_as_load_const(
> > +                     ind->invariant->def->parent_instr);
> > +
> > +         nir_const_value step_val = step_instr->value;
> > +
> > +         nir_load_const_instr *limit_instr =
> > +               nir_instr_as_load_const(
> > +                     limit->nir_loop_var->def->parent_instr);
> > +
> > +         nir_const_value limit_val = limit_instr->value;
> > +
> > +         int iterations = calculate_iterations(&initial_val,
> > &step_val,
> > +                                               &limit_val,
> > cond_op,
> > +                                               ind->alu_def, alu,
> > limit_rhs);
> > +
> > +         /* If this is the first run then overwrite the value */
> > +         if (min_trip_count == -2)
> > +            min_trip_count = iterations;
> > +
> > +         /* Where we not able to calculate the iterations? */
> > +         if (iterations == -1)
> > +            return;
> > +
> > +         /* If we have found a smaller amount of iterations than
> > previously
> > +          * that means we have identified a more limiting
> > terminator
> > +          */
> > +         if (iterations < min_trip_count) {
> > +            min_trip_count = iterations;
> > +            limiting_terminator = terminator;
> > +         }
> > +         break;
> > +
> > +      default:
> > +         return;
> > +      }
> > +   }
> > +
> > +   state->info->is_trip_count_known = true;
> > +   state->info->trip_count = min_trip_count;
> > +   state->info->limiting_terminator = limiting_terminator;
> > +}
> > +
> > +static void
> > +get_loop_info(loop_info_state *state, nir_function_impl *impl)
> > +{
> > +   /* Initialize all variables to "outside_loop". This also marks
> > defs
> > +    * invariant and constant if they are
> > nir_instr_type_load_const's
> > +    */
> > +   nir_foreach_block(block, impl) {
> > +      nir_foreach_instr(instr, block)
> > +         nir_foreach_ssa_def(instr, initialize_ssa_def, state);
> > +   }
> > +
> > +   init_loop_state init_state = {.mark_in_conditional = false,
> > +                                 .mark_nested = false, .state =
> > state };
> > +
> > +   /* Add all entries in the outermost part of the loop to the
> > processing list
> > +    * Mark the entries in conditionals or in nested loops
> > accordingly
> > +    */
> > +   foreach_list_typed_safe(nir_cf_node, node, node, &state->loop-
> > >body) {
> > +      switch (node->type) {
> > +
> > +      case nir_cf_node_block:
> > +         init_state.mark_in_conditional = false;
> > +         init_state.mark_nested = false;
> > +         init_loop_block(nir_cf_node_as_block(node), &init_state);
> > +         break;
> > +
> > +      case nir_cf_node_if:
> > +         init_state.mark_in_conditional = true;
> > +         init_state.mark_nested = false;
> > +         nir_foreach_block_in_cf_node(block, node)
> > +            init_loop_block(block, &init_state);
> > +         break;
> > +
> > +      case nir_cf_node_loop:
> > +         init_state.mark_in_conditional = false;
> > +         init_state.mark_nested = true;
> > +         nir_foreach_block_in_cf_node(block, node)
> > +            init_loop_block(block, &init_state);
> > +         break;
> > +
> > +      case nir_cf_node_function:
> > +         break;
> > +      }
> > +   }
> > +
> > +   /* Induction analysis needs invariance information so get that
> > first */
> > +   compute_invariance_information(state);
> > +
> > +   /* We may now have filled the process_list with instructions
> > from inside
> > +    * the nested blocks in the loop. Remove all instructions from
> > the list
> > +    * nir_foreach_block_in_cf_node before we start computing
> > induction
> > +    * information.
> > +    */
> > +   list_inithead(&state->process_list);
> > +
> > +   /* Add all entries in the outermost part of the loop to the
> > processing list.
> > +    * Don't include defs inn nested loops or in conditionals.
> > +    */
> > +   init_state.mark_in_conditional = false;
> > +   init_state.mark_nested = false;
> > +
> > +   foreach_list_typed_safe(nir_cf_node, node, node, &state->loop-
> > >body)
> > +      if (node->type == nir_cf_node_block)
> > +         init_loop_block(nir_cf_node_as_block(node), &init_state);
> > +
> > +   /* We have invariance information so try to find induction
> > variables */
> > +   if (!compute_induction_information(state))
> > +      return;
> > +
> > +   /* Try to find all simple terminators of the loop. If we can't
> > find any,
> > +    * or we find possible terminators that have side effects then
> > bail.
> > +    */
> > +   if (!find_loop_terminators(state))
> > +      return;
> > +
> > +   /* Run through each of the terminators and try to compute a
> > trip-count */
> > +   find_trip_count(state);
> > +}
> > +
> > +static loop_info_state *
> > +initialize_loop_info_state(nir_loop *loop, void *mem_ctx,
> > nir_function_impl *impl)
> > +{
> > +   loop_info_state *state = rzalloc(mem_ctx, loop_info_state);
> > +   state->loop_vars = rzalloc_array(mem_ctx, loop_variable, impl-
> > >ssa_alloc);
> > +   state->loop = loop;
> > +   state->nir_loop_vars = rzalloc_array(mem_ctx,
> > nir_loop_variable,
> > +                                        impl->ssa_alloc);
> > +
> > +   LIST_INITHEAD(&state->process_list);
> > +
> > +   if (loop->info)
> > +     ralloc_free(loop->info);
> > +
> > +   state->info = rzalloc(loop, nir_loop_info);
> > +
> > +   LIST_INITHEAD(&state->info->loop_terminator_list);
> > +   LIST_INITHEAD(&state->info->loop_vars_list);
> > +
> > +   state->info->var_to_basic_ind =
> > +         _mesa_hash_table_create(state->info, _mesa_hash_pointer,
> > +                                 _mesa_key_pointer_equal);
> > +
> > +   return state;
> > +}
> > +
> > +static void
> > +process_loops(nir_cf_node *cf_node)
> > +{
> > +   switch (cf_node->type) {
> > +   case nir_cf_node_block:
> > +      return;
> > +   case nir_cf_node_if: {
> > +      nir_if *if_stmt = nir_cf_node_as_if(cf_node);
> > +      foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt-
> > >then_list)
> > +         process_loops(nested_node);
> > +      foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt-
> > >else_list)
> > +         process_loops(nested_node);
> > +      return;
> > +   }
> > +   case nir_cf_node_loop: {
> > +      nir_loop *loop = nir_cf_node_as_loop(cf_node);
> > +      foreach_list_typed(nir_cf_node, nested_node, node, &loop-
> > >body)
> > +         process_loops(nested_node);
> > +      break;
> > +   }
> > +   default:
> > +      unreachable("unknown cf node type");
> > +   }
> > +
> > +   nir_loop *loop = nir_cf_node_as_loop(cf_node);
> > +   nir_function_impl *impl = nir_cf_node_get_function(cf_node);
> > +   void *mem_ctx = ralloc_context(NULL);
> > +
> > +   loop_info_state *state = initialize_loop_info_state(loop,
> > mem_ctx, impl);
> > +
> > +   get_loop_info(state, impl);
> > +
> > +   loop->info = state->info;
> > +
> > +   for (int i = 0; i < impl->ssa_alloc; i++) {
> > +      loop_variable *var = &state->loop_vars[i];
> > +      list_add(&var->nir_loop_var->loop_vars_link, &loop->info-
> > >loop_vars_list);
> > +   }
> > +
> > +   ralloc_free(mem_ctx);
> > +}
> > +
> > +void
> > +nir_loop_analyze_impl(nir_function_impl *impl)
> > +{
> > +   if (impl->function->shader->options->max_unroll_iterations ==
> > 0)
> > +      return;
> > +
> > +   nir_index_ssa_defs(impl);
> > +   foreach_list_typed(nir_cf_node, node, node, &impl->body)
> > +      process_loops(node);
> > +}
> > +
> > +void
> > +nir_loop_analyze(nir_shader *shader)
> > +{
> > +   nir_foreach_function(func, shader) {
> > +      if (func->impl)
> > +         nir_loop_analyze_impl(func->impl);
> > +   }
> > +}
> > --
> > 2.7.4
> > 
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list