nir: Remove the old nir_opt_shrink_load.
[mesa.git] / src / compiler / nir / nir.h
index dec20d33fd33e1ece7b9ea6e80648fd21b66d6e2..edae673bfa6d1683a03e97a14d9a2f91c6d4cca7 100644 (file)
@@ -41,6 +41,8 @@
 #include "compiler/nir_types.h"
 #include "compiler/shader_enums.h"
 #include "compiler/shader_info.h"
+#define XXH_INLINE_ALL
+#include "util/xxhash.h"
 #include <stdio.h>
 
 #ifndef NDEBUG
@@ -616,12 +618,52 @@ typedef struct nir_variable {
    struct nir_variable_data *members;
 } nir_variable;
 
-#define nir_foreach_variable(var, var_list) \
+static inline bool
+_nir_shader_variable_has_mode(nir_variable *var, unsigned modes)
+{
+   /* This isn't a shader variable */
+   assert(!(modes & nir_var_function_temp));
+   return var->data.mode & modes;
+}
+
+#define nir_foreach_variable_in_list(var, var_list) \
    foreach_list_typed(nir_variable, var, node, var_list)
 
-#define nir_foreach_variable_safe(var, var_list) \
+#define nir_foreach_variable_in_list_safe(var, var_list) \
    foreach_list_typed_safe(nir_variable, var, node, var_list)
 
+#define nir_foreach_variable_in_shader(var, shader) \
+   nir_foreach_variable_in_list(var, &(shader)->variables)
+
+#define nir_foreach_variable_in_shader_safe(var, shader) \
+   nir_foreach_variable_in_list_safe(var, &(shader)->variables)
+
+#define nir_foreach_variable_with_modes(var, shader, modes) \
+   nir_foreach_variable_in_shader(var, shader) \
+      if (_nir_shader_variable_has_mode(var, modes))
+
+#define nir_foreach_variable_with_modes_safe(var, shader, modes) \
+   nir_foreach_variable_in_shader_safe(var, shader) \
+      if (_nir_shader_variable_has_mode(var, modes))
+
+#define nir_foreach_shader_in_variable(var, shader) \
+   nir_foreach_variable_with_modes(var, shader, nir_var_shader_in)
+
+#define nir_foreach_shader_in_variable_safe(var, shader) \
+   nir_foreach_variable_with_modes_safe(var, shader, nir_var_shader_in)
+
+#define nir_foreach_shader_out_variable(var, shader) \
+   nir_foreach_variable_with_modes(var, shader, nir_var_shader_out)
+
+#define nir_foreach_shader_out_variable_safe(var, shader) \
+   nir_foreach_variable_with_modes_safe(var, shader, nir_var_shader_out)
+
+#define nir_foreach_uniform_variable(var, shader) \
+   nir_foreach_variable_with_modes(var, shader, nir_var_uniform)
+
+#define nir_foreach_uniform_variable_safe(var, shader) \
+   nir_foreach_variable_with_modes_safe(var, shader, nir_var_uniform)
+
 static inline bool
 nir_variable_is_global(const nir_variable *var)
 {
@@ -1021,10 +1063,10 @@ nir_alu_type_get_type_size(nir_alu_type type)
    return type & NIR_ALU_TYPE_SIZE_MASK;
 }
 
-static inline unsigned
+static inline nir_alu_type
 nir_alu_type_get_base_type(nir_alu_type type)
 {
-   return type & NIR_ALU_TYPE_BASE_TYPE_MASK;
+   return (nir_alu_type)(type & NIR_ALU_TYPE_BASE_TYPE_MASK);
 }
 
 static inline nir_alu_type
@@ -1523,6 +1565,7 @@ typedef enum {
 } nir_memory_semantics;
 
 typedef enum {
+   NIR_SCOPE_NONE,
    NIR_SCOPE_INVOCATION,
    NIR_SCOPE_SUBGROUP,
    NIR_SCOPE_WORKGROUP,
@@ -1694,6 +1737,11 @@ typedef enum {
     */
    NIR_INTRINSIC_MEMORY_SCOPE,
 
+   /**
+    * Value of nir_scope.
+    */
+   NIR_INTRINSIC_EXECUTION_SCOPE,
+
    NIR_INTRINSIC_NUM_INDEX_FLAGS,
 
 } nir_intrinsic_index_flag;
@@ -1739,7 +1787,7 @@ typedef struct {
 extern const nir_intrinsic_info nir_intrinsic_infos[nir_num_intrinsics];
 
 static inline unsigned
-nir_intrinsic_src_components(nir_intrinsic_instr *intr, unsigned srcn)
+nir_intrinsic_src_components(const nir_intrinsic_instr *intr, unsigned srcn)
 {
    const nir_intrinsic_info *info = &nir_intrinsic_infos[intr->intrinsic];
    assert(srcn < info->num_srcs);
@@ -1833,6 +1881,7 @@ INTRINSIC_IDX_ACCESSORS(driver_location, DRIVER_LOCATION, unsigned)
 INTRINSIC_IDX_ACCESSORS(memory_semantics, MEMORY_SEMANTICS, nir_memory_semantics)
 INTRINSIC_IDX_ACCESSORS(memory_modes, MEMORY_MODES, nir_variable_mode)
 INTRINSIC_IDX_ACCESSORS(memory_scope, MEMORY_SCOPE, nir_scope)
+INTRINSIC_IDX_ACCESSORS(execution_scope, EXECUTION_SCOPE, nir_scope)
 
 static inline void
 nir_intrinsic_set_align(nir_intrinsic_instr *intrin,
@@ -2223,8 +2272,26 @@ typedef struct {
 } nir_load_const_instr;
 
 typedef enum {
+   /** Return from a function
+    *
+    * This instruction is a classic function return.  It jumps to
+    * nir_function_impl::end_block.  No return value is provided in this
+    * instruction.  Instead, the function is expected to write any return
+    * data to a deref passed in from the caller.
+    */
    nir_jump_return,
+
+   /** Break out of the inner-most loop
+    *
+    * This has the same semantics as C's "break" statement.
+    */
    nir_jump_break,
+
+   /** Jump back to the top of the inner-most loop
+    *
+    * This has the same semantics as C's "continue" statement assuming that a
+    * NIR loop is implemented as "while (1) { body }".
+    */
    nir_jump_continue,
 } nir_jump_type;
 
@@ -2672,6 +2739,15 @@ typedef enum {
     * determine.  Most passes shouldn't preserve this metadata type.
     */
    nir_metadata_loop_analysis = 0x10,
+
+   /** All metadata
+    *
+    * This includes all nir_metadata flags except not_properly_reset.  Passes
+    * which do not change the shader in any way should call
+    *
+    *    nir_metadata_preserve(impl, nir_metadata_all);
+    */
+   nir_metadata_all = ~nir_metadata_not_properly_reset,
 } nir_metadata;
 
 typedef struct {
@@ -2702,6 +2778,12 @@ typedef struct {
    nir_metadata valid_metadata;
 } nir_function_impl;
 
+#define nir_foreach_function_temp_variable(var, impl) \
+   foreach_list_typed(nir_variable, var, node, &(impl)->locals)
+
+#define nir_foreach_function_temp_variable_safe(var, impl) \
+   foreach_list_typed_safe(nir_variable, var, node, &(impl)->locals)
+
 ATTRIBUTE_RETURNS_NONNULL static inline nir_block *
 nir_start_block(nir_function_impl *impl)
 {
@@ -3066,6 +3148,11 @@ typedef struct nir_shader_compiler_options {
    bool vectorize_io;
    bool lower_to_scalar;
 
+   /**
+    * Whether nir_opt_vectorize should only create 16-bit 2D vectors.
+    */
+   bool vectorize_vec2_16bit;
+
    /**
     * Should the linker unify inputs_read/outputs_written between adjacent
     * shader stages which are linked into a single program?
@@ -3102,10 +3189,10 @@ typedef struct nir_shader_compiler_options {
     * to imul with masked inputs and iadd */
    bool has_umad24;
 
-   /* Whether to generate only scoped_memory_barrier intrinsics instead of the
-    * set of memory barrier intrinsics based on GLSL.
+   /* Whether to generate only scoped_barrier intrinsics instead of the set of
+    * memory and control barrier intrinsics based on GLSL.
     */
-   bool use_scoped_memory_barrier;
+   bool use_scoped_barrier;
 
    /**
     * Is this the Intel vec4 backend?
@@ -3134,16 +3221,7 @@ typedef struct nir_shader_compiler_options {
 
 typedef struct nir_shader {
    /** list of uniforms (nir_variable) */
-   struct exec_list uniforms;
-
-   /** list of inputs (nir_variable) */
-   struct exec_list inputs;
-
-   /** list of outputs (nir_variable) */
-   struct exec_list outputs;
-
-   /** list of shared compute variables (nir_variable) */
-   struct exec_list shared;
+   struct exec_list variables;
 
    /** Set of driver-specific options for the shader.
     *
@@ -3155,12 +3233,6 @@ typedef struct nir_shader {
    /** Various bits of compile-time information about a given shader */
    struct shader_info info;
 
-   /** list of global variables in the shader (nir_variable) */
-   struct exec_list globals;
-
-   /** list of system value variables in the shader (nir_variable) */
-   struct exec_list system_values;
-
    struct exec_list functions; /** < list of nir_function */
 
    /**
@@ -3236,6 +3308,14 @@ nir_variable *nir_local_variable_create(nir_function_impl *impl,
                                         const struct glsl_type *type,
                                         const char *name);
 
+nir_variable *nir_find_variable_with_location(nir_shader *shader,
+                                              nir_variable_mode mode,
+                                              unsigned location);
+
+nir_variable *nir_find_variable_with_driver_location(nir_shader *shader,
+                                                     nir_variable_mode mode,
+                                                     unsigned location);
+
 /** creates a function and adds it to the shader's list of functions */
 nir_function *nir_function_create(nir_shader *shader, const char *name);
 
@@ -3253,6 +3333,8 @@ nir_function_impl *nir_cf_node_get_function(nir_cf_node *node);
 void nir_metadata_require(nir_function_impl *impl, nir_metadata required, ...);
 /** dirties all but the preserved metadata */
 void nir_metadata_preserve(nir_function_impl *impl, nir_metadata preserved);
+/** Preserves all metadata for the given shader */
+void nir_shader_preserve_all_metadata(nir_shader *shader);
 
 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
 nir_alu_instr *nir_alu_instr_create(nir_shader *shader, nir_op op);
@@ -3536,6 +3618,9 @@ bool nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb,
                          void *state);
 bool nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state);
 bool nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state);
+bool nir_foreach_phi_src_leaving_block(nir_block *instr,
+                                       nir_foreach_src_cb cb,
+                                       void *state);
 
 nir_const_value *nir_src_as_const_value(nir_src src);
 
@@ -3645,7 +3730,8 @@ unsigned nir_index_instrs(nir_function_impl *impl);
 
 void nir_index_blocks(nir_function_impl *impl);
 
-void nir_index_vars(nir_shader *shader, nir_function_impl *impl, nir_variable_mode modes);
+unsigned nir_shader_index_vars(nir_shader *shader, nir_variable_mode modes);
+unsigned nir_function_impl_index_vars(nir_function_impl *impl);
 
 void nir_print_shader(nir_shader *shader, FILE *fp);
 void nir_print_shader_annotated(nir_shader *shader, FILE *fp, struct hash_table *errors);
@@ -3890,12 +3976,13 @@ void nir_gather_ssa_types(nir_function_impl *impl,
                           BITSET_WORD *float_types,
                           BITSET_WORD *int_types);
 
-void nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
+void nir_assign_var_locations(nir_shader *shader, nir_variable_mode mode,
+                              unsigned *size,
                               int (*type_size)(const struct glsl_type *, bool));
 
 /* Some helpers to do very simple linking */
 bool nir_remove_unused_varyings(nir_shader *producer, nir_shader *consumer);
-bool nir_remove_unused_io_vars(nir_shader *shader, struct exec_list *var_list,
+bool nir_remove_unused_io_vars(nir_shader *shader, nir_variable_mode mode,
                                uint64_t *used_by_other_stage,
                                uint64_t *used_by_other_stage_patches);
 void nir_compact_varyings(nir_shader *producer, nir_shader *consumer,
@@ -3906,7 +3993,8 @@ bool nir_link_opt_varyings(nir_shader *producer, nir_shader *consumer);
 bool nir_lower_amul(nir_shader *shader,
                     int (*type_size)(const struct glsl_type *, bool));
 
-void nir_assign_io_var_locations(struct exec_list *var_list,
+void nir_assign_io_var_locations(nir_shader *shader,
+                                 nir_variable_mode mode,
                                  unsigned *size,
                                  gl_shader_stage stage);
 
@@ -3971,6 +4059,12 @@ typedef enum {
     */
    nir_address_format_32bit_index_offset,
 
+   /**
+    * An address format which is comprised of a vec3 where the first two
+    * components specify the buffer and the third is an offset.
+    */
+   nir_address_format_vec2_index_32bit_offset,
+
    /**
     * An address format which is a simple 32-bit offset.
     */
@@ -3990,12 +4084,13 @@ static inline unsigned
 nir_address_format_bit_size(nir_address_format addr_format)
 {
    switch (addr_format) {
-   case nir_address_format_32bit_global:           return 32;
-   case nir_address_format_64bit_global:           return 64;
-   case nir_address_format_64bit_bounded_global:   return 32;
-   case nir_address_format_32bit_index_offset:     return 32;
-   case nir_address_format_32bit_offset:           return 32;
-   case nir_address_format_logical:                return 32;
+   case nir_address_format_32bit_global:              return 32;
+   case nir_address_format_64bit_global:              return 64;
+   case nir_address_format_64bit_bounded_global:      return 32;
+   case nir_address_format_32bit_index_offset:        return 32;
+   case nir_address_format_vec2_index_32bit_offset:   return 32;
+   case nir_address_format_32bit_offset:              return 32;
+   case nir_address_format_logical:                   return 32;
    }
    unreachable("Invalid address format");
 }
@@ -4004,12 +4099,13 @@ static inline unsigned
 nir_address_format_num_components(nir_address_format addr_format)
 {
    switch (addr_format) {
-   case nir_address_format_32bit_global:           return 1;
-   case nir_address_format_64bit_global:           return 1;
-   case nir_address_format_64bit_bounded_global:   return 4;
-   case nir_address_format_32bit_index_offset:     return 2;
-   case nir_address_format_32bit_offset:           return 1;
-   case nir_address_format_logical:                return 1;
+   case nir_address_format_32bit_global:              return 1;
+   case nir_address_format_64bit_global:              return 1;
+   case nir_address_format_64bit_bounded_global:      return 4;
+   case nir_address_format_32bit_index_offset:        return 2;
+   case nir_address_format_vec2_index_32bit_offset:   return 3;
+   case nir_address_format_32bit_offset:              return 1;
+   case nir_address_format_logical:                   return 1;
    }
    unreachable("Invalid address format");
 }
@@ -4055,7 +4151,8 @@ bool nir_lower_vars_to_ssa(nir_shader *shader);
 
 bool nir_remove_dead_derefs(nir_shader *shader);
 bool nir_remove_dead_derefs_impl(nir_function_impl *impl);
-bool nir_remove_dead_variables(nir_shader *shader, nir_variable_mode modes);
+bool nir_remove_dead_variables(nir_shader *shader, nir_variable_mode modes,
+                               bool (*can_remove_var)(nir_variable *var));
 bool nir_lower_variable_initializers(nir_shader *shader,
                                      nir_variable_mode modes);
 
@@ -4084,6 +4181,7 @@ void nir_lower_io_to_scalar(nir_shader *shader, nir_variable_mode mask);
 void nir_lower_io_to_scalar_early(nir_shader *shader, nir_variable_mode mask);
 bool nir_lower_io_to_vector(nir_shader *shader, nir_variable_mode mask);
 
+bool nir_lower_fragcolor(nir_shader *shader);
 void nir_lower_fragcoord_wtrans(nir_shader *shader);
 void nir_lower_viewport_transform(nir_shader *shader);
 bool nir_lower_uniforms_to_ubo(nir_shader *shader, int multiplier);
@@ -4097,6 +4195,7 @@ typedef struct nir_lower_subgroups_options {
    bool lower_subgroup_masks:1;
    bool lower_shuffle:1;
    bool lower_shuffle_to_32bit:1;
+   bool lower_shuffle_to_swizzle_amd:1;
    bool lower_quad:1;
    bool lower_quad_broadcast_dynamic:1;
    bool lower_quad_broadcast_dynamic_to_const:1;
@@ -4150,6 +4249,8 @@ typedef struct nir_lower_tex_options {
    unsigned lower_xy_uxvx_external;
    unsigned lower_ayuv_external;
    unsigned lower_xyuv_external;
+   unsigned bt709_external;
+   unsigned bt2020_external;
 
    /**
     * To emulate certain texture wrap modes, this can be used
@@ -4310,13 +4411,14 @@ bool nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
 bool nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
                        bool use_clipdist_array);
 bool nir_lower_clip_cull_distance_arrays(nir_shader *nir);
+bool nir_lower_clip_disable(nir_shader *shader, unsigned clip_plane_enable);
 
 void nir_lower_point_size_mov(nir_shader *shader,
                               const gl_state_index16 *pointsize_state_tokens);
 
 bool nir_lower_frexp(nir_shader *nir);
 
-void nir_lower_two_sided_color(nir_shader *shader);
+void nir_lower_two_sided_color(nir_shader *shader, bool face_sysval);
 
 bool nir_lower_clamp_color_outputs(nir_shader *shader);
 
@@ -4383,7 +4485,7 @@ bool nir_lower_bit_size(nir_shader *shader,
                         void *callback_data);
 
 nir_lower_int64_options nir_lower_int64_op_to_options_mask(nir_op opcode);
-bool nir_lower_int64(nir_shader *shader, nir_lower_int64_options options);
+bool nir_lower_int64(nir_shader *shader);
 
 nir_lower_doubles_options nir_lower_doubles_op_to_options_mask(nir_op opcode);
 bool nir_lower_doubles(nir_shader *shader, const nir_shader *softfp64,
@@ -4407,6 +4509,8 @@ bool nir_lower_interpolation(nir_shader *shader,
 
 bool nir_lower_discard_to_demote(nir_shader *shader);
 
+bool nir_lower_memory_model(nir_shader *shader);
+
 bool nir_normalize_cubemap_coords(nir_shader *shader);
 
 void nir_live_ssa_defs_impl(nir_function_impl *impl);
@@ -4514,7 +4618,7 @@ bool nir_opt_rematerialize_compares(nir_shader *shader);
 bool nir_opt_remove_phis(nir_shader *shader);
 bool nir_opt_remove_phis_block(nir_block *block);
 
-bool nir_opt_shrink_load(nir_shader *shader);
+bool nir_opt_shrink_vectors(nir_shader *shader);
 
 bool nir_opt_trivial_continues(nir_shader *shader);
 
@@ -4532,8 +4636,6 @@ bool nir_opt_load_store_vectorize(nir_shader *shader, nir_variable_mode modes,
                                   nir_should_vectorize_mem_func callback,
                                   nir_variable_mode robust_modes);
 
-void nir_schedule(nir_shader *shader, int threshold);
-
 void nir_strip(nir_shader *shader);
 
 void nir_sweep(nir_shader *shader);
@@ -4565,6 +4667,26 @@ nir_variable_is_in_block(const nir_variable *var)
    return nir_variable_is_in_ubo(var) || nir_variable_is_in_ssbo(var);
 }
 
+typedef struct nir_unsigned_upper_bound_config {
+   unsigned min_subgroup_size;
+   unsigned max_subgroup_size;
+   unsigned max_work_group_invocations;
+   unsigned max_work_group_count[3];
+   unsigned max_work_group_size[3];
+
+   uint32_t vertex_attrib_max[32];
+} nir_unsigned_upper_bound_config;
+
+uint32_t
+nir_unsigned_upper_bound(nir_shader *shader, struct hash_table *range_ht,
+                         nir_ssa_scalar scalar,
+                         const nir_unsigned_upper_bound_config *config);
+
+bool
+nir_addition_might_overflow(nir_shader *shader, struct hash_table *range_ht,
+                            nir_ssa_scalar ssa, unsigned const_val,
+                            const nir_unsigned_upper_bound_config *config);
+
 #ifdef __cplusplus
 } /* extern "C" */
 #endif