i965/vec4: Use NIR to do GS input remapping

We're already doing this in the FS back-end.  This just does the same
thing in the vec4 back-end.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

src/intel/compiler/brw_nir.c

@@ -334,7 +334,7 @@ brw_nir_lower_vs_inputs(nir_shader *nir,
 }
 
 void
-brw_nir_lower_vue_inputs(nir_shader *nir, bool is_scalar,
+brw_nir_lower_vue_inputs(nir_shader *nir,
                          const struct brw_vue_map *vue_map)
 {
    foreach_list_typed(nir_variable, var, node, &nir->inputs) {
@@ -344,9 +344,6 @@ brw_nir_lower_vue_inputs(nir_shader *nir, bool is_scalar,
    /* Inputs are stored in vec4 slots, so use type_size_vec4(). */
    nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
 
-   if (nir->stage == MESA_SHADER_GEOMETRY && !is_scalar)
-      return;
-
    /* This pass needs actual constants */
    nir_opt_constant_folding(nir);
 

src/intel/compiler/brw_nir.h

@@ -100,7 +100,7 @@ bool brw_nir_lower_intrinsics(nir_shader *nir,
 void brw_nir_lower_vs_inputs(nir_shader *nir,
                              bool use_legacy_snorm_formula,
                              const uint8_t *vs_attrib_wa_flags);
-void brw_nir_lower_vue_inputs(nir_shader *nir, bool is_scalar,
+void brw_nir_lower_vue_inputs(nir_shader *nir,
                               const struct brw_vue_map *vue_map);
 void brw_nir_lower_tes_inputs(nir_shader *nir, const struct brw_vue_map *vue);
 void brw_nir_lower_fs_inputs(nir_shader *nir,

src/intel/compiler/brw_vec4.cpp

@@ -1677,66 +1677,6 @@ vec4_visitor::dump_instruction(backend_instruction *be_inst, FILE *file)
 }
 
 
-static inline struct brw_reg
-attribute_to_hw_reg(int attr, brw_reg_type type, bool interleaved)
-{
-   struct brw_reg reg;
-
-   unsigned width = REG_SIZE / 2 / MAX2(4, type_sz(type));
-   if (interleaved) {
-      reg = stride(brw_vecn_grf(width, attr / 2, (attr % 2) * 4), 0, width, 1);
-   } else {
-      reg = brw_vecn_grf(width, attr, 0);
-   }
-
-   reg.type = type;
-   return reg;
-}
-
-
-/**
- * Replace each register of type ATTR in this->instructions with a reference
- * to a fixed HW register.
- *
- * If interleaved is true, then each attribute takes up half a register, with
- * register N containing attribute 2*N in its first half and attribute 2*N+1
- * in its second half (this corresponds to the payload setup used by geometry
- * shaders in "single" or "dual instanced" dispatch mode).  If interleaved is
- * false, then each attribute takes up a whole register, with register N
- * containing attribute N (this corresponds to the payload setup used by
- * vertex shaders, and by geometry shaders in "dual object" dispatch mode).
- */
-void
-vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map,
-                                          bool interleaved)
-{
-   foreach_block_and_inst(block, vec4_instruction, inst, cfg) {
-      for (int i = 0; i < 3; i++) {
-         if (inst->src[i].file != ATTR)
-            continue;
-
-         int grf = attribute_map[inst->src[i].nr +
-                                 inst->src[i].offset / REG_SIZE];
-         assert(inst->src[i].offset % REG_SIZE == 0);
-
-         /* All attributes used in the shader need to have been assigned a
-          * hardware register by the caller
-          */
-         assert(grf != 0);
-
-         struct brw_reg reg =
-            attribute_to_hw_reg(grf, inst->src[i].type, interleaved);
-         reg.swizzle = inst->src[i].swizzle;
-         if (inst->src[i].abs)
-            reg = brw_abs(reg);
-         if (inst->src[i].negate)
-            reg = negate(reg);
-
-         inst->src[i] = reg;
-      }
-   }
-}
-
 int
 vec4_vs_visitor::setup_attributes(int payload_reg)
 {

src/intel/compiler/brw_vec4.h

@@ -367,8 +367,6 @@ public:
 
 protected:
    void emit_vertex();
-   void lower_attributes_to_hw_regs(const int *attribute_map,
-                                    bool interleaved);
    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
                                    int reg_node_count);
    virtual void setup_payload() = 0;

src/intel/compiler/brw_vec4_gs_nir.cpp

@@ -66,8 +66,10 @@ vec4_gs_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
       nir_const_value *vertex = nir_src_as_const_value(instr->src[0]);
       nir_const_value *offset_reg = nir_src_as_const_value(instr->src[1]);
 
+      const unsigned input_array_stride = prog_data->urb_read_length * 2;
+
       if (nir_dest_bit_size(instr->dest) == 64) {
-         src = src_reg(ATTR, BRW_VARYING_SLOT_COUNT * vertex->u32[0] +
+         src = src_reg(ATTR, input_array_stride * vertex->u32[0] +
                        instr->const_index[0] + offset_reg->u32[0],
                        glsl_type::dvec4_type);
 
@@ -85,15 +87,11 @@ vec4_gs_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
          /* Make up a type...we have no way of knowing... */
          const glsl_type *const type = glsl_type::ivec(instr->num_components);
 
-         src = src_reg(ATTR, BRW_VARYING_SLOT_COUNT * vertex->u32[0] +
+         src = src_reg(ATTR, input_array_stride * vertex->u32[0] +
                        instr->const_index[0] + offset_reg->u32[0],
                        type);
          src.swizzle = BRW_SWZ_COMP_INPUT(nir_intrinsic_component(instr));
 
-         /* gl_PointSize is passed in the .w component of the VUE header */
-         if (instr->const_index[0] == VARYING_SLOT_PSIZ)
-            src.swizzle = BRW_SWIZZLE_WWWW;
-
          dest = get_nir_dest(instr->dest, src.type);
          dest.writemask = brw_writemask_for_size(instr->num_components);
          emit(MOV(dest, src));

src/intel/compiler/brw_vec4_gs_visitor.cpp

@@ -29,6 +29,7 @@
 
 #include "brw_vec4_gs_visitor.h"
 #include "gen6_gs_visitor.h"
+#include "brw_cfg.h"
 #include "brw_fs.h"
 #include "brw_nir.h"
 #include "common/gen_debug.h"
@@ -72,9 +73,36 @@ vec4_gs_visitor::make_reg_for_system_value(int location)
    return reg;
 }
 
+static inline struct brw_reg
+attribute_to_hw_reg(int attr, brw_reg_type type, bool interleaved)
+{
+   struct brw_reg reg;
 
+   unsigned width = REG_SIZE / 2 / MAX2(4, type_sz(type));
+   if (interleaved) {
+      reg = stride(brw_vecn_grf(width, attr / 2, (attr % 2) * 4), 0, width, 1);
+   } else {
+      reg = brw_vecn_grf(width, attr, 0);
+   }
+
+   reg.type = type;
+   return reg;
+}
+
+/**
+ * Replace each register of type ATTR in this->instructions with a reference
+ * to a fixed HW register.
+ *
+ * If interleaved is true, then each attribute takes up half a register, with
+ * register N containing attribute 2*N in its first half and attribute 2*N+1
+ * in its second half (this corresponds to the payload setup used by geometry
+ * shaders in "single" or "dual instanced" dispatch mode).  If interleaved is
+ * false, then each attribute takes up a whole register, with register N
+ * containing attribute N (this corresponds to the payload setup used by
+ * vertex shaders, and by geometry shaders in "dual object" dispatch mode).
+ */
 int
-vec4_gs_visitor::setup_varying_inputs(int payload_reg, int *attribute_map,
+vec4_gs_visitor::setup_varying_inputs(int payload_reg,
                                       int attributes_per_reg)
 {
    /* For geometry shaders there are N copies of the input attributes, where N
@@ -89,12 +117,24 @@ vec4_gs_visitor::setup_varying_inputs(int payload_reg, int *attribute_map,
    assert(num_input_vertices <= MAX_GS_INPUT_VERTICES);
    unsigned input_array_stride = prog_data->urb_read_length * 2;
 
-   for (int slot = 0; slot < c->input_vue_map.num_slots; slot++) {
+   foreach_block_and_inst(block, vec4_instruction, inst, cfg) {
-      int varying = c->input_vue_map.slot_to_varying[slot];
+      for (int i = 0; i < 3; i++) {
-      for (unsigned vertex = 0; vertex < num_input_vertices; vertex++) {
+         if (inst->src[i].file != ATTR)
-         attribute_map[BRW_VARYING_SLOT_COUNT * vertex + varying] =
+            continue;
-            attributes_per_reg * payload_reg + input_array_stride * vertex +
+
-            slot;
+         assert(inst->src[i].offset % REG_SIZE == 0);
+         int grf = payload_reg * attributes_per_reg +
+                   inst->src[i].nr + inst->src[i].offset / REG_SIZE;
+
+         struct brw_reg reg =
+            attribute_to_hw_reg(grf, inst->src[i].type, attributes_per_reg > 1);
+         reg.swizzle = inst->src[i].swizzle;
+         if (inst->src[i].abs)
+            reg = brw_abs(reg);
+         if (inst->src[i].negate)
+            reg = negate(reg);
+
+         inst->src[i] = reg;
       }
    }
 
@@ -103,25 +143,15 @@ vec4_gs_visitor::setup_varying_inputs(int payload_reg, int *attribute_map,
    return payload_reg + regs_used;
 }
 
-
 void
 vec4_gs_visitor::setup_payload()
 {
-   int attribute_map[BRW_VARYING_SLOT_COUNT * MAX_GS_INPUT_VERTICES];
-
    /* If we are in dual instanced or single mode, then attributes are going
     * to be interleaved, so one register contains two attribute slots.
     */
    int attributes_per_reg =
       prog_data->dispatch_mode == DISPATCH_MODE_4X2_DUAL_OBJECT ? 1 : 2;
 
-   /* If a geometry shader tries to read from an input that wasn't written by
-    * the vertex shader, that produces undefined results, but it shouldn't
-    * crash anything.  So initialize attribute_map to zeros--that ensures that
-    * these undefined results are read from r0.
-    */
-   memset(attribute_map, 0, sizeof(attribute_map));
-
    int reg = 0;
 
    /* The payload always contains important data in r0, which contains
@@ -132,13 +162,11 @@ vec4_gs_visitor::setup_payload()
 
    /* If the shader uses gl_PrimitiveIDIn, that goes in r1. */
    if (gs_prog_data->include_primitive_id)
-      attribute_map[VARYING_SLOT_PRIMITIVE_ID] = attributes_per_reg * reg++;
+      reg++;
 
    reg = setup_uniforms(reg);
 
-   reg = setup_varying_inputs(reg, attribute_map, attributes_per_reg);
+   reg = setup_varying_inputs(reg, attributes_per_reg);
-
-   lower_attributes_to_hw_regs(attribute_map, attributes_per_reg > 1);
 
    this->first_non_payload_grf = reg;
 }
@@ -634,7 +662,7 @@ brw_compile_gs(const struct brw_compiler *compiler, void *log_data,
                        shader->info.separate_shader);
 
    shader = brw_nir_apply_sampler_key(shader, compiler, &key->tex, is_scalar);
-   brw_nir_lower_vue_inputs(shader, is_scalar, &c.input_vue_map);
+   brw_nir_lower_vue_inputs(shader, &c.input_vue_map);
    brw_nir_lower_vue_outputs(shader, is_scalar);
    shader = brw_postprocess_nir(shader, compiler, is_scalar);
 

src/intel/compiler/brw_vec4_gs_visitor.h

@@ -64,8 +64,7 @@ protected:
    virtual void nir_emit_intrinsic(nir_intrinsic_instr *instr);
 
 protected:
-   int setup_varying_inputs(int payload_reg, int *attribute_map,
+   int setup_varying_inputs(int payload_reg, int attributes_per_reg);
-                            int attributes_per_reg);
    void emit_control_data_bits();
    void set_stream_control_data_bits(unsigned stream_id);
 

src/intel/compiler/brw_vec4_tcs.cpp

@@ -413,7 +413,7 @@ brw_compile_tcs(const struct brw_compiler *compiler,
                             nir->info.patch_outputs_written);
 
    nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
-   brw_nir_lower_vue_inputs(nir, is_scalar, &input_vue_map);
+   brw_nir_lower_vue_inputs(nir, &input_vue_map);
    brw_nir_lower_tcs_outputs(nir, &vue_prog_data->vue_map,
                              key->tes_primitive_mode);
    if (key->quads_workaround)

src/intel/compiler/gen6_gs_visitor.cpp

@@ -516,9 +516,7 @@ gen6_gs_visitor::setup_payload()
 
    reg = setup_uniforms(reg);
 
-   reg = setup_varying_inputs(reg, attribute_map, attributes_per_reg);
+   reg = setup_varying_inputs(reg, attributes_per_reg);
-
-   lower_attributes_to_hw_regs(attribute_map, true);
 
    this->first_non_payload_grf = reg;
 }