nir, tu: Add and use load_frag_coord_gmem_ir3

We used load_frag_coord_unscaled_ir3 for loading the fragment coord for
input attachments in GMEM, where the normal scaling for gl_FragCoord
shouldn't be used. However with custom resolve a different scaling will
apply to attachments in GMEM. Separate "unscaled" from "gmem" and rename
the NIR options, in preparation for this.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38451>

src/compiler/nir/nir.h

@@ -6043,8 +6043,8 @@ bool nir_lower_idiv(nir_shader *shader, const nir_lower_idiv_options *options);
 typedef struct nir_input_attachment_options {
    bool use_ia_coord_intrin;
    bool use_view_id_for_layer;
-   bool unscaled_depth_stencil_ir3;
-   uint32_t unscaled_input_attachment_ir3;
+   bool gmem_depth_stencil_ir3;
+   uint32_t gmem_input_attachment_ir3;
 } nir_input_attachment_options;
 
 bool nir_lower_input_attachments(nir_shader *shader,

src/compiler/nir/nir_divergence_analysis.c

@@ -848,6 +848,7 @@ visit_intrinsic(nir_intrinsic_instr *instr, struct divergence_state *state)
    case nir_intrinsic_load_frag_coord_w:
    case nir_intrinsic_load_frag_coord_zw_pan:
    case nir_intrinsic_load_frag_coord_unscaled_ir3:
+   case nir_intrinsic_load_frag_coord_gmem_ir3:
    case nir_intrinsic_load_pixel_coord:
    case nir_intrinsic_load_fully_covered:
    case nir_intrinsic_load_sample_pos:

src/compiler/nir/nir_intrinsics.py

@@ -1509,6 +1509,8 @@ system_value("subgroup_id_shift_ir3", 1)
 # System values for freedreno fragment shaders.
 intrinsic("load_frag_coord_unscaled_ir3", dest_comp=4,
           flags=[CAN_ELIMINATE, CAN_REORDER], bit_sizes=[32])
+intrinsic("load_frag_coord_gmem_ir3", dest_comp=4,
+          flags=[CAN_ELIMINATE, CAN_REORDER], bit_sizes=[32])
 
 # Per-view gl_FragSizeEXT and gl_FragCoord offset.
 intrinsic("load_frag_size_ir3", src_comp=[1], dest_comp=2, indices=[RANGE],

src/compiler/nir/nir_lower_input_attachments.c

@@ -29,21 +29,21 @@ load_frag_coord(nir_builder *b, nir_deref_instr *deref,
                 const nir_input_attachment_options *options)
 {
    nir_def *frag_coord = nir_load_frag_coord(b);
-   if (options->unscaled_input_attachment_ir3 ||
-       options->unscaled_depth_stencil_ir3) {
+   if (options->gmem_input_attachment_ir3 ||
+       options->gmem_depth_stencil_ir3) {
       nir_variable *var = nir_deref_instr_get_variable(deref);
       unsigned base = var->data.index;
-      nir_def *unscaled_frag_coord = nir_load_frag_coord_unscaled_ir3(b);
+      nir_def *gmem_frag_coord = nir_load_frag_coord_gmem_ir3(b);
       if (deref->deref_type == nir_deref_type_array &&
-          options->unscaled_input_attachment_ir3) {
-         nir_def *unscaled =
-            nir_i2b(b, nir_iand(b, nir_ishr(b, nir_imm_int(b, options->unscaled_input_attachment_ir3 >> base), deref->arr.index.ssa),
+          options->gmem_input_attachment_ir3) {
+         nir_def *gmem =
+            nir_i2b(b, nir_iand(b, nir_ishr(b, nir_imm_int(b, options->gmem_input_attachment_ir3 >> base), deref->arr.index.ssa),
                                 nir_imm_int(b, 1)));
-         frag_coord = nir_bcsel(b, unscaled, unscaled_frag_coord, frag_coord);
+         frag_coord = nir_bcsel(b, gmem, gmem_frag_coord, frag_coord);
       } else {
          assert(deref->deref_type == nir_deref_type_var);
-         bool unscaled = base == NIR_VARIABLE_NO_INDEX ? options->unscaled_depth_stencil_ir3 : ((options->unscaled_input_attachment_ir3 >> base) & 1);
-         frag_coord = unscaled ? unscaled_frag_coord : frag_coord;
+         bool gmem = base == NIR_VARIABLE_NO_INDEX ? options->gmem_depth_stencil_ir3 : ((options->gmem_input_attachment_ir3 >> base) & 1);
+         frag_coord = gmem ? gmem_frag_coord : frag_coord;
       }
    }
    return frag_coord;

src/freedreno/vulkan/tu_shader.cc

@@ -1160,6 +1160,7 @@ lower_fdm_filter(const nir_instr *instr, const void *data)
 
    nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
    return intrin->intrinsic == nir_intrinsic_load_frag_size ||
+      intrin->intrinsic == nir_intrinsic_load_frag_coord_gmem_ir3 ||
       (intrin->intrinsic == nir_intrinsic_load_frag_coord &&
        options->adjust_fragcoord);
 }
@@ -1209,6 +1210,10 @@ lower_fdm_instr(struct nir_builder *b, nir_instr *instr, void *data)
                       nir_channel(b, unscaled_coord, 3));
    }
 
+   if (intrin->intrinsic == nir_intrinsic_load_frag_coord_gmem_ir3) {
+      return nir_load_frag_coord_unscaled_ir3(b);
+   }
+
    assert(intrin->intrinsic == nir_intrinsic_load_frag_size);
    return frag_size;
 }
@@ -2771,9 +2776,9 @@ tu_shader_create(struct tu_device *dev,
           * multiview is enabled.
           */
          .use_view_id_for_layer = key->multiview_mask != 0,
-         .unscaled_depth_stencil_ir3 =
+         .gmem_depth_stencil_ir3 =
             key->dynamic_renderpass && !(key->read_only_input_attachments & 1),
-         .unscaled_input_attachment_ir3 =
+         .gmem_input_attachment_ir3 =
             key->dynamic_renderpass ?
             ~(key->read_only_input_attachments >> 1) :
             key->unscaled_input_fragcoord,