[PATCH,i386] Properly check xgetbv for zmm support.
Ilya Tocar
tocarip.intel@gmail.com
Mon Sep 15 15:29:00 GMT 2014
Hi,
Currently we don't check zmm/mask-registers related bits in xgetbv
output, when detecting native cpu. Patch below fixes it.
Bootstraps/passes make check.
Ok for trunk?
ChangeLog:
gcc/
2014-09-15 Ilya Tocar <ilya.tocar@intel.com>
* config/i386/driver-i386.c (host_detect_local_cpu): Detect lack of
zmm/k regs support.
testsuite/
2014-09-15 Ilya Tocar <ilya.tocar@intel.com>
* gcc.target/i386/avx512f-os-support.h: Remove magic number.
---
gcc/config/i386/driver-i386.c | 17 +++++++++++++++++
gcc/testsuite/gcc.target/i386/avx512f-os-support.h | 17 +++++++++++++++--
2 files changed, 32 insertions(+), 2 deletions(-)
diff --git a/gcc/config/i386/driver-i386.c b/gcc/config/i386/driver-i386.c
index af3088e..4d6bf83 100644
--- a/gcc/config/i386/driver-i386.c
+++ b/gcc/config/i386/driver-i386.c
@@ -533,6 +533,9 @@ const char *host_detect_local_cpu (int argc, const char **argv)
#define XSTATE_FP 0x1
#define XSTATE_SSE 0x2
#define XSTATE_YMM 0x4
+#define XSTATE_OPMASK 0x20
+#define XSTATE_ZMM 0x40
+#define XSTATE_HI_ZMM 0x80
if (has_osxsave)
asm (".byte 0x0f; .byte 0x01; .byte 0xd0"
: "=a" (eax), "=d" (edx)
@@ -554,6 +557,20 @@ const char *host_detect_local_cpu (int argc, const char **argv)
has_xsavec = 0;
}
+ if (!has_osxsave
+ || (eax &
+ (XSTATE_SSE | XSTATE_YMM | XSTATE_OPMASK | XSTATE_ZMM | XSTATE_HI_ZMM))
+ != (XSTATE_SSE | XSTATE_YMM | XSTATE_OPMASK | XSTATE_ZMM | XSTATE_HI_ZMM))
+ {
+ has_avx512f = 0;
+ has_avx512er = 0;
+ has_avx512pf = 0;
+ has_avx512cd = 0;
+ has_avx512dq = 0;
+ has_avx512bw = 0;
+ has_avx512vl = 0;
+ }
+
if (!arch)
{
if (vendor == signature_AMD_ebx
diff --git a/gcc/testsuite/gcc.target/i386/avx512f-os-support.h b/gcc/testsuite/gcc.target/i386/avx512f-os-support.h
index deefa5e..2f1ed03 100644
--- a/gcc/testsuite/gcc.target/i386/avx512f-os-support.h
+++ b/gcc/testsuite/gcc.target/i386/avx512f-os-support.h
@@ -1,10 +1,23 @@
/* Check if the OS supports executing AVX512F instructions. */
+#define XCR_XFEATURE_ENABLED_MASK 0x0
+
+#define XSTATE_FP 0x1
+#define XSTATE_SSE 0x2
+#define XSTATE_YMM 0x4
+#define XSTATE_OPMASK 0x20
+#define XSTATE_ZMM 0x40
+#define XSTATE_HI_ZMM 0x80
+
static int
avx512f_os_support (void)
{
unsigned int eax, edx;
+ unsigned int ecx = XCR_XFEATURE_ENABLED_MASK;
+ unsigned int mask = XSTATE_SSE | XSTATE_YMM | XSTATE_OPMASK
+ | XSTATE_ZMM | XSTATE_HI_ZMM;
+
+ __asm__ ("xgetbv" : "=a" (eax), "=d" (edx) : "c" (ecx));
- __asm__ ("xgetbv" : "=a" (eax), "=d" (edx) : "c" (0));
- return (eax & 230) == 230;
+ return ((eax & mask) == mask);
}
--
1.8.3.1
More information about the Gcc-patches
mailing list