* [PATCH] arm: [MVE intrinsics] Fix support for loads [PR target/114323]
@ 2024-03-15 20:08 Christophe Lyon
2024-03-18 17:46 ` Richard Earnshaw (lists)
0 siblings, 1 reply; 2+ messages in thread
From: Christophe Lyon @ 2024-03-15 20:08 UTC (permalink / raw)
To: gcc-patches, richard.sandiford, richard.earnshaw, kyrylo.tkachov
Cc: Christophe Lyon
The testcase in this PR shows that we would load from an uninitialized
location, because the vld1 instrinsics are reported as "const". This
is because function_instance::reads_global_state_p() does not take
CP_READ_MEMORY into account. Fixing this gives vld1 the "pure"
attribute instead, and solves the problem.
2024-03-15 Christophe Lyon <christophe.lyon@linaro.org>
PR target/114323
gcc/
* config/arm/arm-mve-builtins.cc
(function_instance::reads_global_state_p): Take CP_READ_MEMORY
into account.
gcc/testsuite/
* gcc.target/arm/mve/pr114323.c: New.
---
gcc/config/arm/arm-mve-builtins.cc | 2 +-
gcc/testsuite/gcc.target/arm/mve/pr114323.c | 22 +++++++++++++++++++++
2 files changed, 23 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/gcc.target/arm/mve/pr114323.c
diff --git a/gcc/config/arm/arm-mve-builtins.cc b/gcc/config/arm/arm-mve-builtins.cc
index 2f2c0f4a02a..6a5775c67e5 100644
--- a/gcc/config/arm/arm-mve-builtins.cc
+++ b/gcc/config/arm/arm-mve-builtins.cc
@@ -746,7 +746,7 @@ function_instance::reads_global_state_p () const
if (flags & CP_READ_FPCR)
return true;
- return false;
+ return flags & CP_READ_MEMORY;
}
/* Return true if calls to the function could modify some form of
diff --git a/gcc/testsuite/gcc.target/arm/mve/pr114323.c b/gcc/testsuite/gcc.target/arm/mve/pr114323.c
new file mode 100644
index 00000000000..bd9127b886a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/pr114323.c
@@ -0,0 +1,22 @@
+/* { dg-do run } */
+/* { dg-require-effective-target arm_mve_hw } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_1m_mve_fp } */
+
+#include <arm_mve.h>
+
+__attribute__((noipa))
+uint32x4_t foo (void) {
+ uint32x4_t V0 = vld1q_u32(((const uint32_t[4]){1, 2, 3, 4}));
+ return V0;
+}
+
+int main(void)
+{
+ uint32_t buf[4];
+ vst1q_u32 (buf, foo());
+
+ for (int i = 0; i < 4; i++)
+ if (buf[i] != i+1)
+ __builtin_abort ();
+}
--
2.34.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] arm: [MVE intrinsics] Fix support for loads [PR target/114323]
2024-03-15 20:08 [PATCH] arm: [MVE intrinsics] Fix support for loads [PR target/114323] Christophe Lyon
@ 2024-03-18 17:46 ` Richard Earnshaw (lists)
0 siblings, 0 replies; 2+ messages in thread
From: Richard Earnshaw (lists) @ 2024-03-18 17:46 UTC (permalink / raw)
To: Christophe Lyon, gcc-patches, richard.sandiford, kyrylo.tkachov
On 15/03/2024 20:08, Christophe Lyon wrote:
> The testcase in this PR shows that we would load from an uninitialized
> location, because the vld1 instrinsics are reported as "const". This
> is because function_instance::reads_global_state_p() does not take
> CP_READ_MEMORY into account. Fixing this gives vld1 the "pure"
> attribute instead, and solves the problem.
>
> 2024-03-15 Christophe Lyon <christophe.lyon@linaro.org>
>
> PR target/114323
> gcc/
> * config/arm/arm-mve-builtins.cc
> (function_instance::reads_global_state_p): Take CP_READ_MEMORY
> into account.
>
> gcc/testsuite/
> * gcc.target/arm/mve/pr114323.c: New.
OK.
R.
> ---
> gcc/config/arm/arm-mve-builtins.cc | 2 +-
> gcc/testsuite/gcc.target/arm/mve/pr114323.c | 22 +++++++++++++++++++++
> 2 files changed, 23 insertions(+), 1 deletion(-)
> create mode 100644 gcc/testsuite/gcc.target/arm/mve/pr114323.c
>
> diff --git a/gcc/config/arm/arm-mve-builtins.cc b/gcc/config/arm/arm-mve-builtins.cc
> index 2f2c0f4a02a..6a5775c67e5 100644
> --- a/gcc/config/arm/arm-mve-builtins.cc
> +++ b/gcc/config/arm/arm-mve-builtins.cc
> @@ -746,7 +746,7 @@ function_instance::reads_global_state_p () const
> if (flags & CP_READ_FPCR)
> return true;
>
> - return false;
> + return flags & CP_READ_MEMORY;
> }
>
> /* Return true if calls to the function could modify some form of
> diff --git a/gcc/testsuite/gcc.target/arm/mve/pr114323.c b/gcc/testsuite/gcc.target/arm/mve/pr114323.c
> new file mode 100644
> index 00000000000..bd9127b886a
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/mve/pr114323.c
> @@ -0,0 +1,22 @@
> +/* { dg-do run } */
> +/* { dg-require-effective-target arm_mve_hw } */
> +/* { dg-options "-O2" } */
> +/* { dg-add-options arm_v8_1m_mve_fp } */
> +
> +#include <arm_mve.h>
> +
> +__attribute__((noipa))
> +uint32x4_t foo (void) {
> + uint32x4_t V0 = vld1q_u32(((const uint32_t[4]){1, 2, 3, 4}));
> + return V0;
> +}
> +
> +int main(void)
> +{
> + uint32_t buf[4];
> + vst1q_u32 (buf, foo());
> +
> + for (int i = 0; i < 4; i++)
> + if (buf[i] != i+1)
> + __builtin_abort ();
> +}
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-03-18 17:46 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-03-15 20:08 [PATCH] arm: [MVE intrinsics] Fix support for loads [PR target/114323] Christophe Lyon
2024-03-18 17:46 ` Richard Earnshaw (lists)
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).