From 43afcd6bfb1000b4cee976b1bf456c651ab50f74 Mon Sep 17 00:00:00 2001
From: Emil <65846814+emil28092005@users.noreply.github.com>
Date: Wed, 23 Sep 2026 22:40:08 +0300
Subject: [PATCH] Declare R32f HZB image and support chunked GPU culling

---
 shaders/gpu_scene.slang         |  6 +++---
 tests/test_shader_reflection.py | 27 +++++++++++++++++++++++++++
 2 files changed, 30 insertions(+), 3 deletions(-)

diff --git a/shaders/gpu_scene.slang b/shaders/gpu_scene.slang
index 3b07c01..de77059 100644
--- a/shaders/gpu_scene.slang
+++ b/shaders/gpu_scene.slang
@@ -190,7 +190,7 @@ bool appendDeferred(uint candidateIndex) {
 [shader("compute")]
 [numthreads(64, 1, 1)]
 void gpuCullMain(uint3 dispatchId : SV_DispatchThreadID) {
-    uint index = dispatchId.x;
+    uint index = dispatchId.x + cullParameters.reserved0;
     if (index >= cullParameters.candidateCount) return;
     Candidate candidate = candidates[index];
     InstanceRecord instance = cullInstances[candidate.instanceId];
@@ -207,7 +207,7 @@ void gpuCullMain(uint3 dispatchId : SV_DispatchThreadID) {
 [shader("compute")]
 [numthreads(64, 1, 1)]
 void gpuPostCullMain(uint3 dispatchId : SV_DispatchThreadID) {
-    uint index = dispatchId.x;
+    uint index = dispatchId.x + cullParameters.reserved0;
     if (index >= cullParameters.deferredCapacity || index >= deferredCount[0]) return;
     Candidate candidate = candidates[deferredIds[index]];
     InstanceRecord instance = cullInstances[candidate.instanceId];
@@ -224,7 +224,7 @@ struct HzbParameters {
 };
 [[vk::push_constant]] ConstantBuffer<HzbParameters> hzbParameters;
 [[vk::binding(0,0)]] Texture2D<float> hzbSource;
-[[vk::binding(1,0)]] RWTexture2D<float> hzbOutput;
+[[vk::binding(1,0)]] [vk::image_format("r32f")] RWTexture2D<float> hzbOutput;
 [shader("compute")]
 [numthreads(8, 8, 1)]
 void gpuHzbMain(uint3 dispatchId : SV_DispatchThreadID) {
diff --git a/tests/test_shader_reflection.py b/tests/test_shader_reflection.py
index d2caa1c..12eff72 100644
--- a/tests/test_shader_reflection.py
+++ b/tests/test_shader_reflection.py
@@ -76,6 +76,33 @@ class ReflectionTests(unittest.TestCase):
             self.assertEqual([bindings[n]["type"] for n in (7, 8)],
                              ["sampled_image_2d", "sampled_image_2d"])
 
+    def test_hzb_storage_image_has_explicit_r32f_format(self):
+        compiler = os.environ["FASET_TEST_SLANGC"]
+        with tempfile.TemporaryDirectory(prefix="faset-hzb-format-") as directory:
+            process = subprocess.run(
+                [sys.executable, str(SCRIPT), "--compiler", compiler, "--source",
+                 str(SCRIPT.parents[1] / "shaders" / "gpu_scene.slang"), "--entry",
+                 "gpuHzbMain", "--define", "FASET_GPU_HZB=1", "--output", directory],
+                capture_output=True, text=True,
+            )
+            self.assertEqual(process.returncode, 0, process.stderr)
+            bytecode = (Path(directory) / "gpuHzbMain.spv").read_bytes()
+            words = struct.unpack(f"<{len(bytecode) // 4}I", bytecode)
+            capabilities = set()
+            image_formats = []
+            offset = 5
+            while offset < len(words):
+                count, opcode = words[offset] >> 16, words[offset] & 0xffff
+                self.assertGreater(count, 0)
+                if opcode == 17:  # OpCapability
+                    capabilities.add(words[offset + 1])
+                elif opcode == 25:  # OpTypeImage; last operand is ImageFormat
+                    image_formats.append((words[offset + 7], words[offset + 8]))
+                offset += count
+            self.assertNotIn(55, capabilities)  # StorageImageReadWithoutFormat
+            self.assertNotIn(56, capabilities)  # StorageImageWriteWithoutFormat
+            self.assertIn((2, 3), image_formats)  # storage image, R32f
+
 
 if __name__ == "__main__":
     unittest.main()