panfrost: add intrinsic to load frag coord at a barycentric

This is needed for noperspective lowering, where we need to multiply the
varying value by gl_FragCoord.w at the same barycentric as the varying.
Normal nir_load_frag_coord_zw instructions are lowered to the new
intrinsic on bifrost with the pan_lower_frag_coord_zw pass.

Signed-off-by: Benjamin Lee <benjamin.lee@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32127>

src/compiler/nir/nir_divergence_analysis.c

@@ -763,6 +763,7 @@ visit_intrinsic(nir_intrinsic_instr *instr, struct divergence_state *state)
    case nir_intrinsic_load_line_coord:
    case nir_intrinsic_load_frag_coord:
    case nir_intrinsic_load_frag_coord_zw:
+   case nir_intrinsic_load_frag_coord_zw_pan:
    case nir_intrinsic_load_frag_coord_unscaled_ir3:
    case nir_intrinsic_load_pixel_coord:
    case nir_intrinsic_load_fully_covered:

src/compiler/nir/nir_intrinsics.py

@@ -1504,6 +1504,11 @@ store("raw_output_pan", [], [IO_SEMANTICS, BASE])
 store("combined_output_pan", [1, 1, 1, 4], [IO_SEMANTICS, COMPONENT, SRC_TYPE, DEST_TYPE])
 load("raw_output_pan", [1], [IO_SEMANTICS], [CAN_ELIMINATE, CAN_REORDER])
 
+# Like the frag_coord_zw intrinsic, but takes a barycentric. This is needed for
+# noperspective lowering.
+# src[] = { barycoord }
+intrinsic("load_frag_coord_zw_pan", [2], dest_comp=1, indices=[COMPONENT], flags=[CAN_ELIMINATE, CAN_REORDER], bit_sizes=[32])
+
 # Loads the sampler paramaters <min_lod, max_lod, lod_bias>
 # src[] = { sampler_index }
 load("sampler_lod_parameters_pan", [1], flags=[CAN_ELIMINATE, CAN_REORDER])

src/compiler/nir/nir_opt_sink.c

@@ -112,6 +112,7 @@ can_sink_instr(nir_instr *instr, nir_move_options options, bool *can_mov_out_of_
       case nir_intrinsic_load_per_vertex_input:
       case nir_intrinsic_load_frag_coord:
       case nir_intrinsic_load_frag_coord_zw:
+      case nir_intrinsic_load_frag_coord_zw_pan:
       case nir_intrinsic_load_pixel_coord:
       case nir_intrinsic_load_attribute_pan:
          return options & nir_move_load_input;

src/panfrost/compiler/bifrost_compile.c

@@ -1654,11 +1654,21 @@ bi_emit_atomic_i32_to(bi_builder *b, bi_index dst, bi_index addr, bi_index arg,
 }
 
 static void
-bi_emit_load_frag_coord_zw(bi_builder *b, bi_index dst, unsigned channel)
+bi_emit_load_frag_coord_zw_pan(bi_builder *b, nir_intrinsic_instr *instr)
 {
+   bi_index dst = bi_def_index(&instr->def);
+   unsigned channel = nir_intrinsic_component(instr);
+   nir_intrinsic_instr *bary = nir_src_as_intrinsic(instr->src[0]);
+
+   enum bi_sample sample = bi_interp_for_intrinsic(bary->intrinsic);
+   bi_index src0 = bi_varying_src0_for_barycentric(b, bary);
+
+   /* .explicit is not supported with frag_z */
+   if (channel == 2)
+      assert(sample != BI_SAMPLE_EXPLICIT);
+
    bi_ld_var_special_to(
-      b, dst, bi_zero(), BI_REGISTER_FORMAT_F32, BI_SAMPLE_CENTER,
-      BI_UPDATE_CLOBBER,
+      b, dst, src0, BI_REGISTER_FORMAT_F32, sample, BI_UPDATE_CLOBBER,
       (channel == 2) ? BI_VARYING_NAME_FRAG_Z : BI_VARYING_NAME_FRAG_W,
       BI_VECSIZE_NONE);
 }
@@ -1904,8 +1914,8 @@ bi_emit_intrinsic(bi_builder *b, nir_intrinsic_instr *instr)
       bi_mov_i32_to(b, dst, bi_preload(b, 59));
       break;
 
-   case nir_intrinsic_load_frag_coord_zw:
-      bi_emit_load_frag_coord_zw(b, dst, nir_intrinsic_component(instr));
+   case nir_intrinsic_load_frag_coord_zw_pan:
+      bi_emit_load_frag_coord_zw_pan(b, instr);
       break;
 
    case nir_intrinsic_load_converted_output_pan:
@@ -5334,6 +5344,7 @@ bifrost_preprocess_nir(nir_shader *nir, unsigned gpu_id)
    NIR_PASS(_, nir, nir_lower_var_copies);
    NIR_PASS(_, nir, nir_lower_alu);
    NIR_PASS(_, nir, nir_lower_frag_coord_to_pixel_coord);
+   NIR_PASS(_, nir, pan_nir_lower_frag_coord_zw);
 }
 
 static bi_context *

src/panfrost/util/meson.build

@@ -9,6 +9,7 @@ libpanfrost_util_files = files(
   'pan_ir.c',
   'pan_ir.h',
   'pan_liveness.c',
+  'pan_lower_frag_coord_zw.c',
   'pan_lower_framebuffer.c',
   'pan_lower_helper_invocation.c',
   'pan_lower_image_index.c',

src/panfrost/util/pan_ir.h

@@ -386,6 +386,8 @@ bool pan_nir_lower_store_component(nir_shader *shader);
 
 bool pan_nir_lower_image_ms(nir_shader *shader);
 
+bool pan_nir_lower_frag_coord_zw(nir_shader *shader);
+
 bool pan_lower_helper_invocation(nir_shader *shader);
 bool pan_lower_sample_pos(nir_shader *shader);
 bool pan_lower_xfb(nir_shader *nir);

src/panfrost/util/pan_lower_frag_coord_zw.c

@@ -0,0 +1,34 @@
+/*
+ * Copyright © 2024 Collabora Ltd.
+ * SPDX-License-Identifier: MIT
+ */
+
+#include "compiler/nir/nir_builder.h"
+#include "pan_ir.h"
+
+/* Lowers nir_load_frag_coord_zw to nir_load_frag_coord_zw_pan. */
+
+static bool
+lower_frag_coord_zw(nir_builder *b, nir_intrinsic_instr *intrin, void *data)
+{
+   if (intrin->intrinsic != nir_intrinsic_load_frag_coord_zw)
+      return false;
+
+   b->cursor = nir_before_instr(&intrin->instr);
+
+   nir_def *bary = nir_load_barycentric_pixel(b, 32,
+      .interp_mode = INTERP_MODE_NOPERSPECTIVE
+   );
+   unsigned component = nir_intrinsic_component(intrin);
+   nir_def *new = nir_load_frag_coord_zw_pan(b, bary, .component = component);
+   nir_def_replace(&intrin->def, new);
+
+   return true;
+}
+
+bool
+pan_nir_lower_frag_coord_zw(nir_shader *shader)
+{
+   return nir_shader_intrinsics_pass(shader, lower_frag_coord_zw,
+                                     nir_metadata_control_flow, NULL);
+}