agx: Handle fragment shader side effects

Fragment shaders with side effects need to be lowered to ensure they execute for
all shaded pixels but no helper threads. Add a lowering pass to handle this.

Fixes dEQP-GLES31.functional.shaders.opaque_type_indexing.atomic_counter.const_literal_fragment

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21712>

diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index 7f9358c..ca055fb 100644 (file)
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -2239,7 +2239,9 @@ agx_preprocess_nir(nir_shader *nir, bool support_lod_bias)
    NIR_PASS_V(nir, nir_lower_vars_to_ssa);
    NIR_PASS_V(nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
               glsl_type_size, 0);
+   NIR_PASS_V(nir, nir_lower_ssbo);
    if (nir->info.stage == MESA_SHADER_FRAGMENT) {
+      NIR_PASS_V(nir, agx_nir_lower_frag_sidefx);
       NIR_PASS_V(nir, agx_nir_lower_zs_emit);
 
       /* Interpolate varyings at fp16 and write to the tilebuffer at fp16. As an
@@ -2293,7 +2295,6 @@ agx_preprocess_nir(nir_shader *nir, bool support_lod_bias)
    NIR_PASS_V(nir, nir_opt_move, move_all);
    NIR_PASS_V(nir, agx_nir_lower_ubo);
    NIR_PASS_V(nir, agx_nir_lower_shared_bitsize);
-   NIR_PASS_V(nir, nir_lower_ssbo);
 }
 
 void

diff --git a/src/asahi/compiler/agx_compiler.h b/src/asahi/compiler/agx_compiler.h
index 031733e..25ae85d 100644 (file)
--- a/src/asahi/compiler/agx_compiler.h
+++ b/src/asahi/compiler/agx_compiler.h
@@ -834,6 +834,7 @@ bool agx_nir_lower_load_mask(nir_shader *shader);
 bool agx_nir_lower_address(nir_shader *shader);
 bool agx_nir_lower_ubo(nir_shader *shader);
 bool agx_nir_lower_shared_bitsize(nir_shader *shader);
+bool agx_nir_lower_frag_sidefx(nir_shader *s);
 
 #ifdef __cplusplus
 } /* extern C */

diff --git a/src/asahi/compiler/agx_nir_lower_frag_sidefx.c b/src/asahi/compiler/agx_nir_lower_frag_sidefx.c
new file mode 100644 (file)
index 0000000..f261587
--- /dev/null
+++ b/src/asahi/compiler/agx_nir_lower_frag_sidefx.c
@@ -0,0 +1,78 @@
+/*
+ * Copyright 2022 Alyssa Rosenzweig
+ * SPDX-License-Identifier: MIT
+ */
+
+#include "compiler/nir/nir.h"
+#include "compiler/nir/nir_builder.h"
+#include "agx_compiler.h"
+
+/* Fragment shaders with side effects require special handling to ensure the
+ * side effects execute as intended. By default, they require late depth
+ * testing, to ensure the side effects happen even for killed pixels. To handle,
+ * the driver inserts a dummy `gl_FragDepth = gl_Position.z` in shaders that
+ * don't otherwise write their depth, forcing a late depth test.
+ *
+ * For side effects with force early testing forced, the sample mask is written
+ * at the *beginning* of the shader (TODO: handle).
+ */
+
+static bool
+pass(struct nir_builder *b, nir_instr *instr, void *data)
+{
+   if (instr->type != nir_instr_type_intrinsic)
+      return false;
+
+   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
+   if (intr->intrinsic != nir_intrinsic_store_output)
+      return false;
+
+   /* Only lower once */
+   bool *done = data;
+   if (*done)
+      return false;
+   *done = true;
+
+   b->cursor = nir_before_instr(instr);
+   nir_ssa_def *zero = nir_imm_int(b, 0);
+   nir_ssa_def *pixel = nir_load_barycentric_pixel(b, 32, .interp_mode = 1);
+
+   nir_ssa_def *position =
+      nir_load_interpolated_input(b, 1, 32, pixel, zero, .component = 2 /* Z */,
+                                  .dest_type = nir_type_float32,
+                                  .io_semantics = {
+                                     .location = VARYING_SLOT_POS,
+                                     .num_slots = 1,
+                                  });
+
+   nir_store_output(b, position, zero,
+                    .io_semantics.location = FRAG_RESULT_DEPTH,
+                    .write_mask = BITFIELD_MASK(1),
+                    .src_type = nir_type_float32);
+
+   b->shader->info.inputs_read |= BITFIELD64_BIT(VARYING_SLOT_POS);
+   b->shader->info.outputs_written |= BITFIELD64_BIT(FRAG_RESULT_DEPTH);
+   return true;
+}
+
+bool
+agx_nir_lower_frag_sidefx(nir_shader *s)
+{
+   assert(s->info.stage == MESA_SHADER_FRAGMENT);
+
+   /* If there are no side effects, there's nothing to lower */
+   if (!s->info.writes_memory)
+      return false;
+
+   /* Lower writes from helper invocations with the common pass */
+   NIR_PASS_V(s, nir_lower_helper_writes, false);
+
+   /* If depth/stencil feedback is already used, we're done */
+   if (s->info.outputs_written & (BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
+                                  BITFIELD64_BIT(FRAG_RESULT_DEPTH)))
+      return false;
+
+   bool done = false;
+   return nir_shader_instructions_pass(
+      s, pass, nir_metadata_block_index | nir_metadata_dominance, &done);
+}

diff --git a/src/asahi/compiler/meson.build b/src/asahi/compiler/meson.build
index 42af74d..4a751e2 100644 (file)
--- a/src/asahi/compiler/meson.build
+++ b/src/asahi/compiler/meson.build
@@ -25,6 +25,7 @@ libasahi_agx_files = files(
   'agx_liveness.c',
   'agx_insert_waits.c',
   'agx_nir_lower_address.c',
+  'agx_nir_lower_frag_sidefx.c',
   'agx_nir_lower_zs_emit.c',
   'agx_nir_lower_texture.c',
   'agx_nir_lower_load_mask.c',