fix(vulkan): preserve guest mip levels in partial clears

gpu/vk/vk_xfer.cpp

@@ -331,7 +331,8 @@ void clear_tex(VkCommandBuffer cmd, NvnState& st, const nvn_p_clear_tex& p) {
         const nvn_view* v = p.has_view ? &p.view : nullptr;
         for (uint32_t k = 0; k < s.layers; k++) {
             nvn_view lv = v ? *v : nvn_view{};
-            lv.set_levels = 1, lv.base_level = (uint8_t)s.level, lv.nlevels = 1;
+            // texture_view takes a guest mip level and subtracts skip itself.
+            lv.set_levels = 1, lv.base_level = (uint8_t)(s.level + s.tex->skip), lv.nlevels = 1;
             lv.set_layers = 1, lv.min_layer = (uint16_t)(s.layer + k), lv.nlayers = 1;
             const VkImageView view = texture_view(s.tex, p.tex, &lv, VK_IMAGE_VIEW_TYPE_2D, ViewUse::Attachment);
             if (!view) continue;

tests/run_all.py

@@ -30,6 +30,7 @@
     "tests/test_screensaver.py": ["--cc", "gcc"],
     "tests/test_system_header_cache.py": ["--cc", "clang"],
     "tests/test_precompile_compute_guard.py": ["--cxx", "clang++"],
+    "tests/test_vulkan_clear_mip.py": ["--cxx", "{cxx}", "--headers", "{vulkan_headers}"],
     "tests/test_input_background.py": ["--cc", "gcc"],
     "nx2/test_import_windows.py": ["--cxx", "g++"],
     "tests/test_lower_depbar.py": ["--cxx", "clang++"],

tests/test_vulkan_clear_mip.py

@@ -0,0 +1,115 @@
+"""Execute production region resolution and partial clears with instrumented Vulkan commands."""
+import argparse
+from pathlib import Path
+import subprocess
+import tempfile
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def block(source, marker):
+    start = source.index(marker)
+    end = source.index("{", start) + 1
+    depth = 1
+    while depth:
+        depth += (source[end] == "{") - (source[end] == "}")
+        end += 1
+    return source[start:end]
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--cxx", default="clang++")
+    parser.add_argument("--headers", required=True, type=Path)
+    parser.add_argument("--source", type=Path, default=ROOT / "gpu/vk/vk_xfer.cpp")
+    parser.add_argument("--sanitize", action="store_true")
+    args = parser.parse_args()
+    source = args.source.read_text(encoding="utf-8")
+    texture = (ROOT / "gpu/vk/vk_texture.cpp").read_text(encoding="utf-8")
+    fixture = r'''
+#define VK_NO_PROTOTYPES
+#include <vulkan/vulkan_core.h>
+#include <algorithm>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#define CHECK(x) do { if (!(x)) { std::fprintf(stderr,"FAIL:%d %s\n",__LINE__,#x); std::exit(1); } } while (0)
+struct Texture {
+    VkImage image=(VkImage)1; VkImageType type=VK_IMAGE_TYPE_2D;
+    VkImageAspectFlags aspect=VK_IMAGE_ASPECT_COLOR_BIT;
+    uint64_t gpu=0x1000;
+    uint32_t width=64,height=32,depth=1,layers=4,levels=4,skip=0;
+    int nvn_format=1; bool bc=false,astc=false,sparse=false,gpu_written=false,released=false,only_cleared=false;
+} tex;
+struct nvn_tex { uint64_t gpu=0x1000; };
+struct nvn_view { bool set_levels=false,set_layers=false; uint8_t base_level=0,nlevels=0; uint16_t min_layer=0,nlayers=0; };
+struct NVNcopyRegion { int32_t xoffset=0,yoffset=0,zoffset=0,width=0,height=0,depth=1; };
+struct NvnState {};
+union Color { float f[4]; uint32_t u[4]; };
+struct nvn_p_clear_tex { const nvn_tex* tex; nvn_view view{}; bool has_view=true; NVNcopyRegion r{}; uint32_t mask=15; Color c{}; };
+enum class ViewUse { Attachment };
+unsigned selected_mip, expected_mip, first_layer, clear_count, whole_count, touched[4][4];
+void warn(int, const char*) {}
+template<typename... T> void logf(const char*,T...) {}
+Texture* texture_get(const nvn_tex*,bool) { return &tex; }
+int view_nvn_format(const nvn_tex*,const nvn_view*) { return 1; }
+int nvn_block_size(int,int* w,int* h) { *w=*h=1; return 4; }
+void texture_resident(Texture*,uint32_t,uint32_t,uint32_t,uint32_t,uint32_t,uint32_t,uint32_t) {}
+void end_pass(VkCommandBuffer) {}
+void full_barrier(VkCommandBuffer) {}
+struct GprofSpan { GprofSpan(VkCommandBuffer,const char*) {} };
+void texture_written(Texture* t) { t->gpu_written=true; }
+''' + "\n".join(block(texture, m) for m in ("unsigned view_level(", "unsigned image_level(", "unsigned view_layer(")) + r'''
+VkImageView texture_view(Texture* t,const nvn_tex*,const nvn_view* view,VkImageViewType,ViewUse) {
+    selected_mip=image_level(t,view_level(view)); first_layer=view_layer(view);
+    CHECK(view->nlevels==1 && view->nlayers==1);
+    return (VkImageView)1;
+}
+void vkCmdBeginRendering(VkCommandBuffer,const VkRenderingInfo* info) {
+    if (selected_mip != expected_mip) std::fprintf(stderr,"selected mip %u instead of %u\n",selected_mip,expected_mip);
+    CHECK(selected_mip==expected_mip);
+    CHECK(info->renderArea.extent.width==std::max(1u,tex.width>>selected_mip));
+    CHECK(info->renderArea.extent.height==std::max(1u,tex.height>>selected_mip));
+}
+void vkCmdClearAttachments(VkCommandBuffer,uint32_t,const VkClearAttachment*,uint32_t,const VkClearRect*) {
+    ++clear_count; ++touched[selected_mip][first_layer];
+}
+void vkCmdEndRendering(VkCommandBuffer) {}
+void vkCmdClearColorImage(VkCommandBuffer,VkImage,VkImageLayout,const VkClearColorValue*,uint32_t,const VkImageSubresourceRange* r) {
+    CHECK(r->baseMipLevel==expected_mip); ++whole_count;
+}
+void vkCmdClearDepthStencilImage(VkCommandBuffer,VkImage,VkImageLayout,const VkClearDepthStencilValue*,uint32_t,const VkImageSubresourceRange*) { CHECK(false); }
+''' + block(source, "struct Sub {") + ";\n" + block(source, "bool resolve(") + "\n" + block(source, "void clear_tex(") + r'''
+void run(uint32_t skip,uint32_t logical,uint32_t layer,bool whole) {
+    tex=Texture{}; tex.skip=skip; std::memset(touched,0,sizeof touched); clear_count=whole_count=0;
+    expected_mip=logical-skip;
+    nvn_tex input; nvn_p_clear_tex p{&input}; p.view.set_levels=true; p.view.base_level=(uint8_t)logical; p.view.nlevels=1;
+    p.view.set_layers=true; p.view.min_layer=(uint16_t)layer; p.view.nlayers=2;
+    p.r.xoffset=whole ? 0 : 1; p.r.yoffset=whole ? 0 : 1;
+    p.r.width=whole ? (int32_t)(tex.width>>expected_mip) : 2;
+    p.r.height=whole ? (int32_t)(tex.height>>expected_mip) : 2; p.r.depth=2;
+    NvnState state; clear_tex(VK_NULL_HANDLE,state,p);
+    CHECK(tex.gpu_written);
+    CHECK(whole ? whole_count==1 && clear_count==0 : whole_count==0 && clear_count==2);
+    if (!whole) for (unsigned mip=0;mip<4;++mip) for(unsigned l=0;l<4;++l)
+        CHECK(touched[mip][l]==unsigned(mip==expected_mip && l>=layer && l<layer+2));
+}
+int main() {
+    run(1,2,1,false); run(2,3,1,false); run(1,1,0,false);
+    run(0,2,1,false); run(0,0,0,false); run(1,2,1,true);
+    std::puts("PASS: 6 partial/whole clear scenarios, mip mapping and independent array layers");
+}
+'''
+    with tempfile.TemporaryDirectory(prefix="hookshot-clear-mip-") as directory:
+        cpp, exe = Path(directory) / "test.cpp", Path(directory) / "test.exe"
+        cpp.write_text(fixture, encoding="utf-8")
+        flags = ["-fsanitize=address,undefined", "-fno-sanitize-recover=all"] if args.sanitize else []
+        subprocess.run([args.cxx,"-std=c++20","-O1","-Wall","-Werror",
+                        "-Wno-missing-field-initializers",*flags,
+                        "-I"+str(args.headers),str(cpp),"-o",str(exe)],check=True)
+        subprocess.run([str(exe)],check=True,timeout=10)
+
+
+if __name__ == "__main__":
+    main()