[Mesa-dev] [PATCH 10/23] i965/vs: Simplify fs_visitor's ATTR file.

[Kenneth Graunke]

Previously, ATTR was indexed by VERT_ATTRIB_* slots; at the end of
compilation, assign_vs_urb_setup() translated those into GRF units,
and converted ATTR to HW_REGs.

This patch moves the transslation earlier, making ATTR work in terms of
GRF units from the beginning.  assign_vs_urb_setup() simply has to add
the number of payload registers and push constants to obtain the final
hardware GRF number.  (We can't do this earlier as those values aren't
known.)

ATTR still supports reg_offset; however, it's simply added to reg.
It's not clear whether this is valuable or not.

Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
---
 src/mesa/drivers/dri/i965/brw_fs.cpp         | 27 +++++--------------
 src/mesa/drivers/dri/i965/brw_fs_visitor.cpp |  2 +-
 src/mesa/drivers/dri/i965/brw_nir.c          | 40 ++++++++++++++++++++++++++++
 3 files changed, 48 insertions(+), 21 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index e23cb18..76ceb5e 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -1522,9 +1522,11 @@ void
 fs_visitor::assign_vs_urb_setup()
 {
    brw_vs_prog_data *vs_prog_data = (brw_vs_prog_data *) prog_data;
-   int grf, slot, channel, attr;
 
    assert(stage == MESA_SHADER_VERTEX);
+   int count = _mesa_bitcount_64(vs_prog_data->inputs_read);
+   if (vs_prog_data->uses_vertexid || vs_prog_data->uses_instanceid)
+      count++;
 
    /* Each attribute is 4 regs. */
    this->first_non_payload_grf += 4 * vs_prog_data->nr_attributes;
@@ -1535,25 +1537,10 @@ fs_visitor::assign_vs_urb_setup()
    foreach_block_and_inst(block, fs_inst, inst, cfg) {
       for (int i = 0; i < inst->sources; i++) {
          if (inst->src[i].file == ATTR) {
-
-            if (inst->src[i].reg == VERT_ATTRIB_MAX) {
-               slot = vs_prog_data->nr_attributes - 1;
-            } else {
-               /* Attributes come in in a contiguous block, ordered by their
-                * gl_vert_attrib value.  That means we can compute the slot
-                * number for an attribute by masking out the enabled
-                * attributes before it and counting the bits.
-                */
-               attr = inst->src[i].reg + inst->src[i].reg_offset / 4;
-               slot = _mesa_bitcount_64(vs_prog_data->inputs_read &
-                                        BITFIELD64_MASK(attr));
-            }
-
-            channel = inst->src[i].reg_offset & 3;
-
-            grf = payload.num_regs +
-               prog_data->curb_read_length +
-               slot * 4 + channel;
+            int grf = payload.num_regs +
+                      prog_data->curb_read_length +
+                      inst->src[i].reg +
+                      inst->src[i].reg_offset;
 
             inst->src[i].file = HW_REG;
             inst->src[i].fixed_hw_reg =
diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
index 47d7ae4..cac9fff 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
@@ -53,7 +53,7 @@ fs_reg *
 fs_visitor::emit_vs_system_value(int location)
 {
    fs_reg *reg = new(this->mem_ctx)
-      fs_reg(ATTR, VERT_ATTRIB_MAX, BRW_REGISTER_TYPE_D);
+      fs_reg(ATTR, 4*_mesa_bitcount_64(prog->InputsRead), BRW_REGISTER_TYPE_D);
    brw_vs_prog_data *vs_prog_data = (brw_vs_prog_data *) prog_data;
 
    switch (location) {
diff --git a/src/mesa/drivers/dri/i965/brw_nir.c b/src/mesa/drivers/dri/i965/brw_nir.c
index 56bd3d3..a5e8e14 100644
--- a/src/mesa/drivers/dri/i965/brw_nir.c
+++ b/src/mesa/drivers/dri/i965/brw_nir.c
@@ -27,6 +27,34 @@
 #include "glsl/nir/glsl_to_nir.h"
 #include "program/prog_to_nir.h"
 
+static bool
+remap_vs_attrs(nir_block *block, void *closure)
+{
+   GLbitfield64 inputs_read = *((GLbitfield64 *) closure);
+
+   nir_foreach_instr(block, instr) {
+      if (instr->type != nir_instr_type_intrinsic)
+         continue;
+
+      nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+
+      /* We set EmitNoIndirect for VS inputs, so there are no indirects. */
+      assert(intrin->intrinsic != nir_intrinsic_load_input_indirect);
+
+      if (intrin->intrinsic == nir_intrinsic_load_input) {
+         /* Attributes come in a contiguous block, ordered by their
+          * gl_vert_attrib value.  That means we can compute the slot
+          * number for an attribute by masking out the enabled attributes
+          * before it and counting the bits.
+          */
+         int attr = intrin->const_index[0];
+         int slot = _mesa_bitcount_64(inputs_read & BITFIELD64_MASK(attr));
+         intrin->const_index[0] = 4 * slot;
+      }
+   }
+   return true;
+}
+
 static void
 brw_nir_lower_inputs(nir_shader *nir,
                      const struct gl_program *prog,
@@ -51,6 +79,18 @@ brw_nir_lower_inputs(nir_shader *nir,
        * type_size_vec4 here.
        */
       nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+
+      /* Finally, translate VERT_ATTRIB_* values into the actual registers.
+       *
+       * Note that we can use prog->InputsRead rather than key->inputs_read,
+       * since the two identical aside from Gen4-5 edge flag differences.
+       */
+      GLbitfield64 inputs_read = prog->InputsRead;
+      nir_foreach_overload(nir, overload) {
+         if (overload->impl) {
+            nir_foreach_block(overload->impl, remap_vs_attrs, &inputs_read);
+         }
+      }
       break;
    case MESA_SHADER_GEOMETRY:
       foreach_list_typed(nir_variable, var, node, &nir->inputs) {
-- 
2.5.3