[PATCH v3 0/2] MR7908: Fix macOS winetest timeout in kernel32:virtual

[Jinoh Kang (＠iamahuman)]

Work around slow runtime of current FlushProcessWriteBuffers() implementation on macOS.

The total iteration count becomes:

```
- positive half: 32768 * 1 (old) = 8192 * 4 (new)
- negative half: 32768 * 1 (old) > 8192 * 1 (new)
```

--
  v3: kernel32/tests: Shorten time for negative half of litmus test for FlushProcessWriteBuffers().
      kernel32/tests: Run store_buffer_litmus_test() in a single-iteration loop.

https://gitlab.winehq.org/wine/wine/-/merge_requests/7908

[PATCH v3 1/2] kernel32/tests: Run store_buffer_litmus_test() in a single-iteration loop.

[Jinoh Kang]

From: Jinoh Kang

Prepare for a refactoring to vary iteration count for each test.
---
 dlls/kernel32/tests/virtual.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/dlls/kernel32/tests/virtual.c b/dlls/kernel32/tests/virtual.c
index c08c82dd999..07ccd3111d7 100644
--- a/dlls/kernel32/tests/virtual.c
+++ b/dlls/kernel32/tests/virtual.c
@@ -4721,10 +4721,18 @@ static void test_FlushProcessWriteBuffers(void)
         return;
     }

-    reorderings = store_buffer_litmus_test( compiler_barrier, compiler_barrier );
+    /* NOTE: Avoid passing the loop count directly as a parameter to ensure
+     * NOTE: that each test is independent of the differing loop count.
+     */
+
+    reorderings = 0;
+    for (i = 0; i < 1; i++)
+        reorderings += store_buffer_litmus_test( compiler_barrier, compiler_barrier );
     ok( reorderings, "expected write-read reordering with compiler barrier only (got %ld reorderings)\n", reorderings );

-    reorderings = store_buffer_litmus_test( compiler_barrier, pFlushProcessWriteBuffers );
+    reorderings = 0;
+    for (i = 0; i < 1; i++)
+        reorderings += store_buffer_litmus_test( compiler_barrier, pFlushProcessWriteBuffers );
     ok( !reorderings, "expected sequential consistency with FlushProcessWriteBuffers (got %ld reorderings)\n", reorderings );
 }

[PATCH v3 2/2] kernel32/tests: Shorten time for negative half of litmus test for FlushProcessWriteBuffers().

[Jinoh Kang]

From: Jinoh Kang

Work around slow runtime of current FlushProcessWriteBuffers()
implementation on macOS.

The total iteration count becomes:

- positive half: 32768 * 1 (old) = 4096 * 8 (new)
- negative half: 32768 * 1 (old) > 4096 * 2 (new)
---
 dlls/kernel32/tests/virtual.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/dlls/kernel32/tests/virtual.c b/dlls/kernel32/tests/virtual.c
index 07ccd3111d7..7f1715207b5 100644
--- a/dlls/kernel32/tests/virtual.c
+++ b/dlls/kernel32/tests/virtual.c
@@ -4655,7 +4655,7 @@ static LONG store_buffer_litmus_test( void (*WINAPI barrier0)(void), void (*WINA
         .num_slots = 64,

/* Increase if flaky, decrease if slow */
-        .num_generations = 32768,
+        .num_generations = 4096,

.reorderings = &reorderings,
     };
@@ -4726,12 +4726,12 @@ static void test_FlushProcessWriteBuffers(void)
      */

reorderings = 0;
-    for (i = 0; i < 1; i++)
+    for (i = 0; i < 8; i++)  /* run multiple times to reduce flakiness */
         reorderings += store_buffer_litmus_test( compiler_barrier, compiler_barrier );
     ok( reorderings, "expected write-read reordering with compiler barrier only (got %ld reorderings)\n", reorderings );

reorderings = 0;
-    for (i = 0; i < 1; i++)
+    for (i = 0; i < 2; i++)  /* TODO: FlushProcessWriteBuffers() is slow on macOS */
         reorderings += store_buffer_litmus_test( compiler_barrier, pFlushProcessWriteBuffers );
     ok( !reorderings, "expected sequential consistency with FlushProcessWriteBuffers (got %ld reorderings)\n", reorderings );
 }