From 665938d0b157ff1dce056b5c24a57aee9392929b Mon Sep 17 00:00:00 2001
From: snowstormxd <214507868+snowstormxd@users.noreply.github.com>
Date: Mon, 2 Nov 2026 18:00:14 +0000
Subject: macro: execute native parameters without a second FIFO copy

Use the owned parameter vector for DMA FIFO appends. Add a native JIT regression covering reallocation and repeated invocations against the copied-input path.

Assisted by Codex.

(cherry picked from commit 5b6803aaab18e332493903ee49cab026c041d5dc)
---
 src/tests/video_core/native_macro_fifo.cpp | 74 ++++++++++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 src/tests/video_core/native_macro_fifo.cpp

(limited to 'src/tests/video_core/native_macro_fifo.cpp')

diff --git a/src/tests/video_core/native_macro_fifo.cpp b/src/tests/video_core/native_macro_fifo.cpp
new file mode 100644
index 0000000000..e19b146192
--- /dev/null
+++ b/src/tests/video_core/native_macro_fifo.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <array>
+#include <vector>
+#include <catch2/catch_test_macros.hpp>
+
+#ifdef ARCHITECTURE_x86_64
+#include "video_core/macro_turing_jit.h"
+
+namespace {
+std::array<u32, 3> FifoInstruction(u32 method, bool end) {
+    std::array<u32, 3> bits{};
+    const auto field = [&](u32 offset, u32 count, u32 value) {
+        for (u32 bit = 0; bit < count; ++bit) {
+            bits[(offset + bit) / 32] |= ((value >> bit) & 1) << ((offset + bit) % 32);
+        }
+    };
+    field(0, 1, end);
+    for (u32 lane = 0; lane < 2; ++lane) {
+        field(15 + lane * 36, 5, 24);
+        field(20 + lane * 36, 5, 24);
+        field(25 + lane * 36, 5, 24);
+    }
+    field(30, 16, method);
+    field(82, 3, 5);
+    field(85, 4, 3);
+    return {bits[2], bits[1], bits[0]};
+}
+} // namespace
+
+TEST_CASE("Owned native macro FIFOs preserve loads across DMA appends and repeated calls",
+          "[video_core][native_macro]") {
+    std::vector<u32> code;
+    for (const auto instruction : {FifoInstruction(0x158, false), FifoInstruction(0x200, true),
+                                   FifoInstruction(0x201, false)}) {
+        code.insert(code.end(), instruction.begin(), instruction.end());
+    }
+    Tegra::TuringMacroJit jit;
+    jit.Prepare(code, 1);
+    std::array<u32, 0xc00> ram{};
+    std::vector<u32> parameters;
+    std::vector<u32> reference;
+    for (u32 invocation = 0; invocation < 1000; ++invocation) {
+        const u32 count = 2 + invocation % 127;
+        parameters.push_back(count);
+        reference.assign(parameters.begin(), parameters.end());
+        const auto run = [&](std::vector<u32>& input) {
+            std::vector<std::pair<u32, u32>> writes;
+            const auto result = jit.Execute(
+                0, input, ram, [](u32) { return 0U; },
+                [&](u32 method, u32 value) {
+                    writes.emplace_back(method, value);
+                    if (method == 0x158) {
+                        for (u32 word = 0; word < value; ++word) {
+                            input.push_back(invocation * 256 + word);
+                        }
+                    }
+                    return true;
+                });
+            CHECK(result.success);
+            CHECK(result.consumed == 3);
+            return writes;
+        };
+        const auto expected = run(reference);
+        const auto actual = run(parameters);
+        REQUIRE(actual.size() == 3);
+        CHECK(actual == expected);
+        CHECK(parameters == reference);
+        CHECK(actual[1] == std::pair<u32, u32>{0x200, invocation * 256});
+        CHECK(actual[2] == std::pair<u32, u32>{0x201, invocation * 256 + 1});
+        parameters.clear();
+    }
+}
+#endif
-- 
cgit v1.2.3