v3d/compiler: Implement load_output

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33766>

src/broadcom/compiler/meson.build

@@ -24,6 +24,7 @@ libbroadcom_compiler_files = files(
   'v3d_nir_lower_logic_ops.c',
   'v3d_nir_lower_scratch.c',
   'v3d_nir_lower_txf_ms.c',
+  'v3d_nir_lower_load_output.c',
   'v3d_packing.c',
 )
 

src/broadcom/compiler/v3d_compiler.h

@@ -1211,6 +1211,9 @@ bool v3d_nir_lower_image_load_store(nir_shader *s, struct v3d_compile *c);
 bool v3d_nir_lower_global_2x32(nir_shader *s);
 bool v3d_nir_lower_load_store_bitsize(nir_shader *s);
 bool v3d_nir_lower_algebraic(struct nir_shader *shader, const struct v3d_compile *c);
+bool v3d_nir_lower_load_output(nir_shader *s, struct v3d_compile *c);
+
+nir_def *v3d_nir_get_tlb_color(nir_builder *b, struct v3d_compile *c, int rt, int sample);
 
 void v3d_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr);
 void v3d_vir_emit_image_load_store(struct v3d_compile *c,

src/broadcom/compiler/v3d_nir_lower_load_output.c

@@ -0,0 +1,63 @@
+
+#include "util/format/u_format.h"
+#include "compiler/nir/nir_builder.h"
+#include "v3d_compiler.h"
+
+nir_def *
+v3d_nir_get_tlb_color(nir_builder *b, struct v3d_compile *c, int rt, int sample)
+{
+        uint32_t num_components =
+                util_format_get_nr_components(c->fs_key->color_fmt[rt].format);
+
+        nir_def *color[4];
+        for (int i = 0; i < num_components; i++) {
+                color[i] =
+                        nir_load_tlb_color_brcm(b, 1, 32,
+                                                nir_imm_int(b, rt),
+                                                .base = sample,
+                                                .component = i);
+        }
+        return nir_pad_vec4(b, nir_vec(b, color, num_components));
+}
+
+static bool
+lower_load_output(nir_builder *b, nir_intrinsic_instr *intr, void *data)
+{
+        if (intr->intrinsic != nir_intrinsic_load_output)
+                return false;
+
+        b->cursor = nir_before_instr(&intr->instr);
+
+        struct v3d_compile *c = data;
+
+        nir_io_semantics sem = nir_intrinsic_io_semantics(intr);
+        unsigned rt = sem.location - FRAG_RESULT_DATA0;
+
+        unsigned num_samples = c->fs_key->msaa ? V3D_MAX_SAMPLES : 1;
+        assert(num_samples > 0);
+
+        nir_def *sample_id = num_samples > 1 ? nir_load_sample_id(b) : NULL;
+        nir_def *out = v3d_nir_get_tlb_color(b, c, rt, 0);
+
+        /* If msaa, run through every sample and, if it matches the current sample
+         * id, use it.
+         */
+
+        for (unsigned i = 1; i < num_samples; i++) {
+                nir_def *is_cur_sample = nir_ieq_imm(b, sample_id, i);
+                nir_def *val = v3d_nir_get_tlb_color(b, c, rt, i);
+                out = nir_bcsel(b, is_cur_sample, val, out);
+        }
+
+        nir_def_replace(&intr->def, out);
+
+        return true;
+}
+
+bool
+v3d_nir_lower_load_output(nir_shader *s, struct v3d_compile *c)
+{
+        return  nir_shader_intrinsics_pass(s, lower_load_output,
+                                           nir_metadata_control_flow,
+                                           c);
+}

src/broadcom/compiler/v3d_nir_lower_logic_ops.c

@@ -201,28 +201,6 @@ v3d_get_format_swizzle_for_rt(struct v3d_compile *c, int rt)
         }
 }
 
-static nir_def *
-v3d_nir_get_tlb_color(nir_builder *b, struct v3d_compile *c, int rt, int sample)
-{
-        uint32_t num_components =
-                util_format_get_nr_components(c->fs_key->color_fmt[rt].format);
-
-        nir_def *color[4];
-        for (int i = 0; i < 4; i++) {
-                if (i < num_components) {
-                        color[i] =
-                                nir_load_tlb_color_brcm(b, 1, 32,
-                                                        nir_imm_int(b, rt),
-                                                        .base = sample,
-                                                        .component = i);
-                } else {
-                        /* These will be DCEd */
-                        color[i] = nir_imm_int(b, 0);
-                }
-        }
-        return nir_vec4(b, color[0], color[1], color[2], color[3]);
-}
-
 static nir_def *
 v3d_emit_logic_op_raw(struct v3d_compile *c, nir_builder *b,
                       nir_def **src_chans, nir_def **dst_chans,

src/broadcom/compiler/vir.c

@@ -1136,6 +1136,7 @@ v3d_nir_lower_fs_early(struct v3d_compile *c)
         if (c->fs_key->int_color_rb || c->fs_key->uint_color_rb)
                 v3d_fixup_fs_output_types(c);
 
+        NIR_PASS(_, c->s, v3d_nir_lower_load_output, c);
         NIR_PASS(_, c->s, v3d_nir_lower_logic_ops, c);
 
         if (c->fs_key->line_smoothing) {