nir: Add nir_address_format_null_value()
[mesa.git] / src / compiler / nir / nir.h
index 11f9b396e5676b88de9f085331c63210f1daa6ad..7e51501e9130a050e825ec2d69a8ef7e013ad9e2 100644 (file)
@@ -862,8 +862,27 @@ nir_get_nir_type_for_glsl_type(const struct glsl_type *type)
 nir_op nir_type_conversion_op(nir_alu_type src, nir_alu_type dst,
                               nir_rounding_mode rnd);
 
+static inline nir_op
+nir_op_vec(unsigned components)
+{
+   switch (components) {
+   case  1: return nir_op_imov;
+   case  2: return nir_op_vec2;
+   case  3: return nir_op_vec3;
+   case  4: return nir_op_vec4;
+   default: unreachable("bad component count");
+   }
+}
+
 typedef enum {
-   NIR_OP_IS_COMMUTATIVE = (1 << 0),
+   /**
+    * Operation where the first two sources are commutative.
+    *
+    * For 2-source operations, this just mathematical commutativity.  Some
+    * 3-source operations, like ffma, are only commutative in the first two
+    * sources.
+    */
+   NIR_OP_IS_2SRC_COMMUTATIVE = (1 << 0),
    NIR_OP_IS_ASSOCIATIVE = (1 << 1),
 } nir_op_algebraic_property;
 
@@ -1041,20 +1060,7 @@ typedef struct {
    nir_dest dest;
 } nir_deref_instr;
 
-NIR_DEFINE_CAST(nir_instr_as_deref, nir_instr, nir_deref_instr, instr,
-                type, nir_instr_type_deref)
-
-static inline nir_deref_instr *
-nir_src_as_deref(nir_src src)
-{
-   if (!src.is_ssa)
-      return NULL;
-
-   if (src.ssa->parent_instr->type != nir_instr_type_deref)
-      return NULL;
-
-   return nir_instr_as_deref(src.ssa->parent_instr);
-}
+static inline nir_deref_instr *nir_src_as_deref(nir_src src);
 
 static inline nir_deref_instr *
 nir_deref_instr_parent(const nir_deref_instr *instr)
@@ -1457,7 +1463,8 @@ typedef enum {
    nir_texop_txl,                /**< Texture look-up with explicit LOD */
    nir_texop_txd,                /**< Texture look-up with partial derivatives */
    nir_texop_txf,                /**< Texel fetch with explicit LOD */
-   nir_texop_txf_ms,                /**< Multisample texture fetch */
+   nir_texop_txf_ms,             /**< Multisample texture fetch */
+   nir_texop_txf_ms_fb,          /**< Multisample texture fetch from framebuffer */
    nir_texop_txf_ms_mcs,         /**< Multisample compression value fetch */
    nir_texop_txs,                /**< Texture size */
    nir_texop_lod,                /**< Texture lod query */
@@ -1592,6 +1599,7 @@ nir_tex_instr_is_query(const nir_tex_instr *instr)
    case nir_texop_txd:
    case nir_texop_txf:
    case nir_texop_txf_ms:
+   case nir_texop_txf_ms_fb:
    case nir_texop_tg4:
       return false;
    default:
@@ -1631,6 +1639,7 @@ nir_tex_instr_src_type(const nir_tex_instr *instr, unsigned src)
       switch (instr->op) {
       case nir_texop_txf:
       case nir_texop_txf_ms:
+      case nir_texop_txf_ms_fb:
       case nir_texop_txf_ms_mcs:
       case nir_texop_samples_identical:
          return nir_type_int;
@@ -1793,6 +1802,8 @@ typedef struct {
 
 NIR_DEFINE_CAST(nir_instr_as_alu, nir_instr, nir_alu_instr, instr,
                 type, nir_instr_type_alu)
+NIR_DEFINE_CAST(nir_instr_as_deref, nir_instr, nir_deref_instr, instr,
+                type, nir_instr_type_deref)
 NIR_DEFINE_CAST(nir_instr_as_call, nir_instr, nir_call_instr, instr,
                 type, nir_instr_type_call)
 NIR_DEFINE_CAST(nir_instr_as_jump, nir_instr, nir_jump_instr, instr,
@@ -2138,6 +2149,21 @@ nir_loop_last_block(nir_loop *loop)
    return nir_cf_node_as_block(exec_node_data(nir_cf_node, tail, node));
 }
 
+/**
+ * Return true if this list of cf_nodes contains a single empty block.
+ */
+static inline bool
+nir_cf_list_is_empty_block(struct exec_list *cf_list)
+{
+   if (exec_list_is_singular(cf_list)) {
+      struct exec_node *head = exec_list_get_head(cf_list);
+      nir_block *block =
+         nir_cf_node_as_block(exec_node_data(nir_cf_node, head, node));
+      return exec_list_is_empty(&block->instr_list);
+   }
+   return false;
+}
+
 typedef struct {
    uint8_t num_components;
    uint8_t bit_size;
@@ -2242,6 +2268,9 @@ typedef struct nir_shader_compiler_options {
    /** enables rules to lower isign to imin+imax */
    bool lower_isign;
 
+   /** enables rules to lower fsign to fsub and flt */
+   bool lower_fsign;
+
    /* Does the native fdot instruction replicate its result for four
     * components?  If so, then opt_algebraic_late will turn all fdotN
     * instructions into fdot_replicatedN instructions.
@@ -2278,12 +2307,6 @@ typedef struct nir_shader_compiler_options {
    bool lower_all_io_to_temps;
    bool lower_all_io_to_elements;
 
-   /**
-    * Does the driver support real 32-bit integers?  (Otherwise, integers
-    * are simulated by floats.)
-    */
-   bool native_integers;
-
    /* Indicates that the driver only has zero-based vertex id */
    bool vertex_id_zero_based;
 
@@ -2307,6 +2330,14 @@ typedef struct nir_shader_compiler_options {
     */
    bool lower_helper_invocation;
 
+   /**
+    * Convert gl_SampleMaskIn to gl_HelperInvocation as follows:
+    *
+    *   gl_SampleMaskIn == 0 ---> gl_HelperInvocation
+    *   gl_SampleMaskIn != 0 ---> !gl_HelperInvocation
+    */
+   bool optimize_sample_mask_in;
+
    bool lower_cs_local_index_from_id;
    bool lower_cs_local_id_from_index;
 
@@ -2740,27 +2771,18 @@ bool nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state);
 
 nir_const_value *nir_src_as_const_value(nir_src src);
 
-static inline struct nir_instr *
-nir_src_instr(const struct nir_src *src)
-{
-   return src->is_ssa ? src->ssa->parent_instr : NULL;
-}
-
 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro)                \
 static inline c_type *                                                  \
-nir_src_as_ ## name (struct nir_src *src)                               \
-{                                                                       \
-    return src->is_ssa && src->ssa->parent_instr->type == type_enum     \
-           ? cast_macro(src->ssa->parent_instr) : NULL;                 \
-}                                                                       \
-static inline const c_type *                                            \
-nir_src_as_ ## name ## _const(const struct nir_src *src)                \
+nir_src_as_ ## name (nir_src src)                                       \
 {                                                                       \
-    return src->is_ssa && src->ssa->parent_instr->type == type_enum     \
-           ? cast_macro(src->ssa->parent_instr) : NULL;                 \
+    return src.is_ssa && src.ssa->parent_instr->type == type_enum       \
+           ? cast_macro(src.ssa->parent_instr) : NULL;                  \
 }
 
 NIR_SRC_AS_(alu_instr, nir_alu_instr, nir_instr_type_alu, nir_instr_as_alu)
+NIR_SRC_AS_(intrinsic, nir_intrinsic_instr,
+            nir_instr_type_intrinsic, nir_instr_as_intrinsic)
+NIR_SRC_AS_(deref, nir_deref_instr, nir_instr_type_deref, nir_instr_as_deref)
 
 bool nir_src_is_dynamically_uniform(nir_src src);
 bool nir_srcs_equal(nir_src src1, nir_src src2);
@@ -3035,6 +3057,10 @@ bool nir_lower_vars_to_scratch(nir_shader *shader,
 
 void nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint);
 
+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,
                               int (*type_size)(const struct glsl_type *, bool));
 
@@ -3086,7 +3112,70 @@ typedef enum {
     * component is a buffer index and the second is an offset.
     */
    nir_address_format_32bit_index_offset,
+
+   /**
+    * An address format which is a simple 32-bit offset.
+    */
+   nir_address_format_32bit_offset,
+
+   /**
+    * An address format representing a purely logical addressing model.  In
+    * this model, all deref chains must be complete from the dereference
+    * operation to the variable.  Cast derefs are not allowed.  These
+    * addresses will be 32-bit scalars but the format is immaterial because
+    * you can always chase the chain.
+    */
+   nir_address_format_logical,
 } nir_address_format;
+
+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;
+   }
+   unreachable("Invalid address format");
+}
+
+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;
+   }
+   unreachable("Invalid address format");
+}
+
+static inline const struct glsl_type *
+nir_address_format_to_glsl_type(nir_address_format addr_format)
+{
+   unsigned bit_size = nir_address_format_bit_size(addr_format);
+   assert(bit_size == 32 || bit_size == 64);
+   return glsl_vector_type(bit_size == 32 ? GLSL_TYPE_UINT : GLSL_TYPE_UINT64,
+                           nir_address_format_num_components(addr_format));
+}
+
+const nir_const_value *nir_address_format_null_value(nir_address_format addr_format);
+
+nir_ssa_def * nir_explicit_io_address_from_deref(struct nir_builder *b,
+                                                 nir_deref_instr *deref,
+                                                 nir_ssa_def *base_addr,
+                                                 nir_address_format addr_format);
+void nir_lower_explicit_io_instr(struct nir_builder *b,
+                                 nir_intrinsic_instr *io_instr,
+                                 nir_ssa_def *addr,
+                                 nir_address_format addr_format);
+
 bool nir_lower_explicit_io(nir_shader *shader,
                            nir_variable_mode modes,
                            nir_address_format);
@@ -3112,9 +3201,14 @@ bool nir_lower_vec_to_movs(nir_shader *shader);
 void nir_lower_alpha_test(nir_shader *shader, enum compare_func func,
                           bool alpha_to_one);
 bool nir_lower_alu(nir_shader *shader);
-bool nir_lower_alu_to_scalar(nir_shader *shader);
+
+bool nir_lower_flrp(nir_shader *shader, unsigned lowering_mask,
+                    bool always_precise, bool have_ffma);
+
+bool nir_lower_alu_to_scalar(nir_shader *shader, BITSET_WORD *lower_set);
 bool nir_lower_bool_to_float(nir_shader *shader);
 bool nir_lower_bool_to_int32(nir_shader *shader);
+bool nir_lower_int_to_float(nir_shader *shader);
 bool nir_lower_load_const_to_scalar(nir_shader *shader);
 bool nir_lower_read_invocation_to_scalar(nir_shader *shader);
 bool nir_lower_phis_to_scalar(nir_shader *shader);
@@ -3125,6 +3219,7 @@ 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);
 
+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);
 
@@ -3230,6 +3325,12 @@ typedef struct nir_lower_tex_options {
     */
    unsigned lower_srgb;
 
+   /**
+    * If true, lower nir_texop_tex on shaders that doesn't support implicit
+    * LODs to nir_texop_txl.
+    */
+   bool lower_tex_without_implicit_lod;
+
    /**
     * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
     */
@@ -3271,6 +3372,12 @@ typedef struct nir_lower_tex_options {
     */
    bool lower_txd_offset_clamp;
 
+   /**
+    * If true, lower nir_texop_txd with min_lod to a nir_texop_txl if the
+    * sampler is bindless.
+    */
+   bool lower_txd_clamp_bindless_sampler;
+
    /**
     * If true, lower nir_texop_txd with min_lod to a nir_texop_txl if the
     * sampler index is not statically determinable to be less than 16.
@@ -3332,6 +3439,8 @@ bool nir_lower_wpos_ytransform(nir_shader *shader,
                                const nir_lower_wpos_ytransform_options *options);
 bool nir_lower_wpos_center(nir_shader *shader, const bool for_sample_shading);
 
+bool nir_lower_fb_read(nir_shader *shader);
+
 typedef struct nir_lower_drawpixels_options {
    gl_state_index16 texcoord_state_tokens[STATE_LENGTH];
    gl_state_index16 scale_state_tokens[STATE_LENGTH];
@@ -3472,6 +3581,8 @@ uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs, uint64_t dual_slot);
 nir_intrinsic_op nir_intrinsic_from_system_value(gl_system_value val);
 gl_system_value nir_system_value_from_intrinsic(nir_intrinsic_op intrin);
 
+bool nir_lower_sincos(nir_shader *shader);
+
 #ifdef __cplusplus
 } /* extern "C" */
 #endif