nir: add iabs-lowering code

[mesa.git] / src / compiler / nir / nir.h

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 5804670fb9902ee31b6bb087043b945dab97a692..aa7ff44985e98f3896242352ff23aa7a9ae223ef 100644 (file)
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -328,7 +328,7 @@ typedef struct nir_variable {
        *
        * \sa nir_variable_mode
        */
-      nir_variable_mode mode:11;
+      unsigned mode:11;
 
       /**
        * Is the variable read-only?
@@ -467,12 +467,12 @@ typedef struct nir_variable {
       unsigned per_view:1;
 
       /**
-       * \brief Layout qualifier for gl_FragDepth.
+       * \brief Layout qualifier for gl_FragDepth. See nir_depth_layout.
        *
        * This is not equal to \c ir_depth_layout_none if and only if this
        * variable is \c gl_FragDepth and a layout qualifier is specified.
        */
-      nir_depth_layout depth_layout:3;
+      unsigned depth_layout:3;
 
       /**
        * Vertex stream output identifier.
@@ -483,10 +483,12 @@ typedef struct nir_variable {
       unsigned stream:9;
 
       /**
+       * See gl_access_qualifier.
+       *
        * Access flags for memory variables (SSBO/global), image uniforms, and
        * bindless images in uniforms/inputs/outputs.
        */
-      enum gl_access_qualifier access:8;
+      unsigned access:8;
 
       /**
        * Descriptor set binding for sampler or UBO.
@@ -1380,7 +1382,7 @@ nir_alu_instr_is_comparison(const nir_alu_instr *instr)
    case nir_op_flt:
    case nir_op_fge:
    case nir_op_feq:
-   case nir_op_fne:
+   case nir_op_fneu:
    case nir_op_ilt:
    case nir_op_ult:
    case nir_op_ige:
@@ -1852,6 +1854,12 @@ nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val)                \
    const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];   \
    assert(info->index_map[NIR_INTRINSIC_##flag] > 0);                         \
    instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val;       \
+}                                                                             \
+static inline bool                                                            \
+nir_intrinsic_has_##name(nir_intrinsic_instr *instr)                          \
+{                                                                             \
+   const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];   \
+   return info->index_map[NIR_INTRINSIC_##flag] > 0;                          \
 }
 
 INTRINSIC_IDX_ACCESSORS(write_mask, WRMASK, unsigned)
@@ -3016,7 +3024,7 @@ typedef struct nir_shader_compiler_options {
    /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
    bool lower_sub;
 
-   /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
+   /* lower {slt,sge,seq,sne} to {flt,fge,feq,fneu} + b2f: */
    bool lower_scmp;
 
    /* lower fall_equalN/fany_nequalN (ex:fany_nequal4 to sne+fdot4+fsat) */
@@ -3034,6 +3042,9 @@ typedef struct nir_shader_compiler_options {
    /** enables rules to lower fsign to fsub and flt */
    bool lower_fsign;
 
+   /** enables rules to lower iabs to ineg+imax */
+   bool lower_iabs;
+
    /* lower fdph to fdot4 */
    bool lower_fdph;
 
@@ -3116,6 +3127,9 @@ typedef struct nir_shader_compiler_options {
    bool lower_cs_local_index_from_id;
    bool lower_cs_local_id_from_index;
 
+   /* Prevents lowering global_invocation_id to be in terms of work_group_id */
+   bool has_cs_global_id;
+
    bool lower_device_index_to_zero;
 
    /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
@@ -4265,6 +4279,14 @@ bool nir_lower_subgroups(nir_shader *shader,
 
 bool nir_lower_system_values(nir_shader *shader);
 
+typedef struct nir_lower_compute_system_values_options {
+   bool has_base_global_invocation_id:1;
+   bool has_base_work_group_id:1;
+} nir_lower_compute_system_values_options;
+
+bool nir_lower_compute_system_values(nir_shader *shader,
+                                     const nir_lower_compute_system_values_options *options);
+
 enum PACKED nir_lower_tex_packing {
    nir_lower_tex_packing_none = 0,
    /* The sampler returns up to 2 32-bit words of half floats or 16-bit signed