On Thu, May 28, 2026 at 12:50 AM oltolm <[email protected]> wrote:
>
> On x86_64-w64-mingw32, PR target/54412 is triggered when GCC creates
> under-aligned stack storage for AVX values that are passed, returned or
> materialized indirectly.
>
> Add coverage for the original by-value cases, for isolated hidden sret
> allocation, for callee-side by-reference parameter setup, for the reduced
> real-world C++ temporary-slot reproducer, and for an aligned(64) AVX512
> case.
>
> gcc/testsuite/ChangeLog:
>
> PR target/54412
> * gcc.target/i386/pr54412-v4d-o0-aligned-locals.c: New test.
> * gcc.target/i386/pr54412-o2-by-value-cases.c: New test.
> * gcc.target/i386/pr54412-sret-no-args.c: New test.
> * gcc.target/i386/pr54412-callee-byref-param.c: New test.
> * gcc.target/i386/pr54412-avx512-aligned64.c: New test.
> * g++.target/i386/pr54412-setactivepreset.C: New test.
> * g++.target/i386/pr54412-setactivepreset-2.C: New test.
>
> Signed-off-by: oltolm <[email protected]>
> ---
> .../i386/pr54412-setactivepreset-2.C | 58 ++++++++++
> .../g++.target/i386/pr54412-setactivepreset.C | 45 ++++++++
> .../i386/pr54412-avx512-aligned64.c | 29 +++++
> .../i386/pr54412-callee-byref-param.c | 27 +++++
> .../i386/pr54412-o2-by-value-cases.c | 104 ++++++++++++++++++
> .../gcc.target/i386/pr54412-sret-no-args.c | 36 ++++++
> .../i386/pr54412-v4d-o0-aligned-locals.c | 27 +++++
> 7 files changed, 326 insertions(+)
> create mode 100644 gcc/testsuite/g++.target/i386/pr54412-setactivepreset-2.C
> create mode 100644 gcc/testsuite/g++.target/i386/pr54412-setactivepreset.C
> create mode 100644 gcc/testsuite/gcc.target/i386/pr54412-avx512-aligned64.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr54412-callee-byref-param.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr54412-o2-by-value-cases.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr54412-sret-no-args.c
> create mode 100644
> gcc/testsuite/gcc.target/i386/pr54412-v4d-o0-aligned-locals.c
>
> diff --git a/gcc/testsuite/g++.target/i386/pr54412-setactivepreset-2.C
> b/gcc/testsuite/g++.target/i386/pr54412-setactivepreset-2.C
> new file mode 100644
> index 00000000000..4f0abec6ec0
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/pr54412-setactivepreset-2.C
> @@ -0,0 +1,58 @@
> +/* PR target/54412 */
> +/* { dg-do compile { target x86_64-*-mingw* } } */
> +/* { dg-options "-O2 -mavx2 -masm=intel -std=gnu++17
> -fno-omit-frame-pointer" } */
> +/* { dg-final { scan-assembler-not {lea[ \t]+r8, 63\[rsp\]} } } */
> +/* { dg-final { scan-assembler-not {and[ \t]+r8, -32} } } */
> +/* { dg-final { scan-assembler-not {vmovdqa[ \t]+YMMWORD PTR \[r8\], ymm0} }
> } */
> +
> +template <typename _InputIterator, typename _Predicate>
> +__attribute__ ((noinline, noclone))
> +_InputIterator
> +find_if_inner (_InputIterator first, _InputIterator last, _Predicate &pred)
> +{
> + asm volatile ("" : : "g" (first), "g" (last), "g" (&pred) : "memory");
> + return first;
> +}
> +
> +template <typename _InputIterator, typename _Predicate>
> +__attribute__ ((noinline, noclone))
> +_InputIterator
> +find_if (_InputIterator first, _InputIterator last, _Predicate pred)
> +{
> + return find_if_inner (first, last, pred);
> +}
> +
> +struct basic_string_view
> +{
> + int n;
> + void *_r;
> +};
> +
> +struct dummy_iter
> +{
> + int n;
> +};
> +
> +dummy_iter first;
> +dummy_iter last;
> +basic_string_view cit_name;
> +basic_string_view cit_category;
> +
> +struct ic_
> +{
> + void
> + set_active_preset (basic_string_view category,
> + basic_string_view name,
> + bool)
> + {
> + find_if (first, last, [category, name] (dummy_iter) { return false; });
> + }
> +};
> +
> +ic_ cit_pack;
> +
> +void
> +it (bool update_visibility)
> +{
> + cit_pack.set_active_preset (cit_category, cit_name, update_visibility);
> +}
> diff --git a/gcc/testsuite/g++.target/i386/pr54412-setactivepreset.C
> b/gcc/testsuite/g++.target/i386/pr54412-setactivepreset.C
> new file mode 100644
> index 00000000000..c0b989fdb2c
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/pr54412-setactivepreset.C
> @@ -0,0 +1,45 @@
> +/* PR target/54412 */
> +/* { dg-do compile { target x86_64-*-mingw* } } */
> +/* { dg-options "-O2 -mavx2 -masm=intel -std=gnu++17
> -fno-omit-frame-pointer" } */
> +/* { dg-final { scan-assembler-not {vmovdqa[ \t]+YMMWORD PTR -[0-9]+\[rbp\],
> ymm[0-9]+} } } */
> +
> +template <typename _InputIterator, typename _Predicate>
> +__attribute__ ((noinline, noclone))
> +void
> +find_if (_InputIterator first, _InputIterator last, _Predicate pred)
> +{
> + asm volatile ("" : : "g" (first), "g" (last), "g" (&pred) : "memory");
> +}
> +
> +struct basic_string_view
> +{
> + int n;
> + void *_r;
> +
> + int cbegin () const;
> + int cend () const;
> +};
> +
> +basic_string_view presets;
> +basic_string_view cit_name;
> +basic_string_view cit_category;
> +
> +struct ic_
> +{
> + void
> + set_active_preset (basic_string_view category,
> + basic_string_view name,
> + bool)
> + {
> + find_if (&basic_string_view::cbegin, &basic_string_view::cend,
> + [category, name] {});
> + }
> +};
> +
> +ic_ cit_pack;
> +
> +void
> +it (bool update_visibility)
> +{
> + cit_pack.set_active_preset (cit_category, cit_name, update_visibility);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr54412-avx512-aligned64.c
> b/gcc/testsuite/gcc.target/i386/pr54412-avx512-aligned64.c
> new file mode 100644
> index 00000000000..c496f76e76e
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54412-avx512-aligned64.c
> @@ -0,0 +1,29 @@
> +/* PR target/54412 */
> +/* { dg-do run { target { avx512f && { ! ia32 } } } } */
> +/* { dg-skip-if "PR target/54412 exercises the Win64 ABI" { !
> x86_64-*-mingw* } } */
Drop this. Execution test should work for all.
> +/* { dg-options "-O0 -mavx512f" } */
> +
> +#include "avx512-check.h"
> +
> +typedef double v8d __attribute__ ((vector_size (64), aligned (64)));
> +
> +__attribute__ ((noinline)) static v8d
> +roundtrip_v8d (v8d x)
> +{
> + return x;
> +}
> +
> +static void
> +test_512 (void)
> +{
> + v8d x __attribute__ ((aligned (64)))
> + = { 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0 };
> + v8d expected __attribute__ ((aligned (64)))
> + = { 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0 };
> + v8d r __attribute__ ((aligned (64)));
> +
> + r = roundtrip_v8d (x);
> +
> + if (__builtin_memcmp (&r, &expected, sizeof (r)) != 0)
> + __builtin_abort ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr54412-callee-byref-param.c
> b/gcc/testsuite/gcc.target/i386/pr54412-callee-byref-param.c
> new file mode 100644
> index 00000000000..1eb1164ff9b
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54412-callee-byref-param.c
> @@ -0,0 +1,27 @@
> +/* PR target/54412 */
> +/* { dg-do run { target { avx && { ! ia32 } } } } */
> +/* { dg-skip-if "PR target/54412 exercises the Win64 ABI" { !
> x86_64-*-mingw* } } */
Drop this. Execution test should work for all.
> +/* { dg-options "-O0 -mavx" } */
> +
> +#include "avx-check.h"
> +
> +typedef float v8f __attribute__ ((vector_size (32), aligned (32)));
> +
> +static const v8f expected = { 1.0f, 2.0f, 3.0f, 4.0f,
> + 5.0f, 6.0f, 7.0f, 8.0f };
> +
> +__attribute__ ((noinline)) static int
> +param_matches (v8f x)
> +{
> + return __builtin_memcmp (&x, &expected, sizeof (x)) == 0;
> +}
> +
> +static void
> +avx_test (void)
> +{
> + v8f x __attribute__ ((aligned (32)))
> + = { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f };
> +
> + if (!param_matches (x))
> + __builtin_abort ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr54412-o2-by-value-cases.c
> b/gcc/testsuite/gcc.target/i386/pr54412-o2-by-value-cases.c
> new file mode 100644
> index 00000000000..cdcca7b31f4
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54412-o2-by-value-cases.c
> @@ -0,0 +1,104 @@
> +/* PR target/54412 */
> +/* { dg-do run { target { avx && { ! ia32 } } } } */
> +/* { dg-skip-if "PR target/54412 exercises the Win64 ABI" { !
> x86_64-*-mingw* } } */
Drop this. Execution test should work for all.
> +/* { dg-options "-O2 -mavx -std=gnu99" } */
> +
> +#include "avx-check.h"
> +#include <immintrin.h>
> +
> +typedef double v4d __attribute__ ((vector_size (32), aligned (32)));
> +typedef char byte_vec __attribute__ ((vector_size (32)));
> +
> +struct avx_wrapper
> +{
> + __m256 value;
> +};
> +
> +#define PERM_LEFT_BVEC \
> + (byte_vec) { 63, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, \
> + 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, \
> + 29, 30 }
> +
> +__attribute__ ((noinline, noclone)) static v4d
> +roundtrip_v4d (v4d x)
> +{
> + return x;
> +}
> +
> +__attribute__ ((noinline)) static __m256
> +pass_m256 (__m256 x)
> +{
> + return x;
> +}
> +
> +__attribute__ ((noinline)) static struct avx_wrapper
> +pass_wrapper (struct avx_wrapper x)
> +{
> + return x;
> +}
> +
> +__attribute__ ((noinline)) static byte_vec
> +permute (byte_vec x, byte_vec y)
> +{
> + return __builtin_shuffle (x, y, PERM_LEFT_BVEC);
> +}
> +
> +static void
> +test_v4d_roundtrip (void)
> +{
> + v4d x = (v4d) { 1.0, 2.0, 3.0, 4.0 };
> + v4d r = roundtrip_v4d (x);
> + v4d expected = (v4d) { 1.0, 2.0, 3.0, 4.0 };
> +
> + if (__builtin_memcmp (&r, &expected, sizeof (r)) != 0)
> + __builtin_abort ();
> +}
> +
> +static void
> +test_m256_and_wrapper_by_value (void)
> +{
> + __m256 m = _mm256_set_ps (8.0f, 7.0f, 6.0f, 5.0f,
> + 4.0f, 3.0f, 2.0f, 1.0f);
> + __m256 r = pass_m256 (m);
> + struct avx_wrapper x;
> + struct avx_wrapper y;
> +
> + x.value = _mm256_set_ps (16.0f, 15.0f, 14.0f, 13.0f,
> + 12.0f, 11.0f, 10.0f, 9.0f);
> + y = pass_wrapper (x);
> +
> + if (__builtin_memcmp (&r, &m, sizeof (r)) != 0)
> + __builtin_abort ();
> +
> + if (__builtin_memcmp (&y, &x, sizeof (y)) != 0)
> + __builtin_abort ();
> +}
> +
> +static void
> +test_two_arg_bytevec_shuffle (void)
> +{
> + byte_vec x = { 1, 2, 3, 4, 5, 6, 7, 8,
> + 9, 10, 11, 12, 13, 14, 15, 16,
> + 17, 18, 19, 20, 21, 22, 23, 24,
> + 25, 26, 27, 28, 29, 30, 31, 32 };
> + byte_vec y = { 11, 12, 13, 14, 15, 16, 17, 18,
> + 19, 110, 111, 112, 113, 114, 115, 116,
> + 117, 118, 119, 120, 121, 122, 123, 124,
> + 125, 126, 127, 88, 89, 90, 91, 92 };
> + byte_vec expected = { 92, 1, 2, 3, 4, 5, 6, 7,
> + 8, 9, 10, 11, 12, 13, 14, 15,
> + 16, 17, 18, 19, 20, 21, 22, 23,
> + 24, 25, 26, 27, 28, 29, 30, 31 };
> + byte_vec z = permute (x, y);
> +
> + if (__builtin_memcmp (&z, &expected, sizeof (z)) != 0)
> + __builtin_abort ();
> +}
> +
> +static void
> +avx_test (void)
> +{
> + test_v4d_roundtrip ();
> + test_m256_and_wrapper_by_value ();
> + test_two_arg_bytevec_shuffle ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr54412-sret-no-args.c
> b/gcc/testsuite/gcc.target/i386/pr54412-sret-no-args.c
> new file mode 100644
> index 00000000000..af23e6a99a0
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54412-sret-no-args.c
> @@ -0,0 +1,36 @@
> +/* PR target/54412 */
> +/* { dg-do run { target { avx && { ! ia32 } } } } */
> +/* { dg-skip-if "PR target/54412 exercises the Win64 ABI" { !
> x86_64-*-mingw* } } */
Drop this. Execution test should work for all.
> +/* { dg-options "-O2 -mavx -std=gnu99" } */
> +
> +#include "avx-check.h"
> +#include <immintrin.h>
> +
> +struct avx_wrapper
> +{
> + __m256 value;
> +};
> +
> +__attribute__ ((noinline)) static struct avx_wrapper
> +make_wrapper (void)
> +{
> + struct avx_wrapper x;
> +
> + x.value = _mm256_set_ps (8.0f, 7.0f, 6.0f, 5.0f,
> + 4.0f, 3.0f, 2.0f, 1.0f);
> + return x;
> +}
> +
> +static void
> +avx_test (void)
> +{
> + struct avx_wrapper expected;
> + struct avx_wrapper r;
> +
> + expected.value = _mm256_set_ps (8.0f, 7.0f, 6.0f, 5.0f,
> + 4.0f, 3.0f, 2.0f, 1.0f);
> + r = make_wrapper ();
> +
> + if (__builtin_memcmp (&r, &expected, sizeof (r)) != 0)
> + __builtin_abort ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr54412-v4d-o0-aligned-locals.c
> b/gcc/testsuite/gcc.target/i386/pr54412-v4d-o0-aligned-locals.c
> new file mode 100644
> index 00000000000..74675f53bcb
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54412-v4d-o0-aligned-locals.c
> @@ -0,0 +1,27 @@
> +/* PR target/54412 */
> +/* { dg-do run { target { avx && { ! ia32 } } } } */
Why !ia32? Why do these tests fail for ia32?
> +/* { dg-skip-if "PR target/54412 exercises the Win64 ABI" { !
> x86_64-*-mingw* } } */
Drop this. Execution test should work for all.
> +/* { dg-options "-O0 -mavx" } */
> +
> +#include "avx-check.h"
> +
> +typedef double v4d __attribute__ ((vector_size (32)));
> +
> +__attribute__ ((noinline)) static v4d
> +f (v4d x)
> +{
> + return x;
> +}
> +
> +static void
> +avx_test (void)
> +{
> + v4d x __attribute__ ((aligned (32))) = { 1.0, 2.0, 3.0, 4.0 };
> + v4d expected __attribute__ ((aligned (32))) = { 1.0, 2.0, 3.0, 4.0 };
> + v4d r __attribute__ ((aligned (32)));
> +
> + r = f (x);
> +
> + if (__builtin_memcmp (&r, &expected, sizeof (r)) != 0)
> + __builtin_abort ();
> +}
> --
> 2.54.0.windows.1
>
--
H.J.
