Patchwork [x86-64,BZ,#20024] Fixed vector sincos/sincosf ABI

login
register
mail settings
Submitter Andrew Senkevich
Date July 1, 2016, 11:53 a.m.
Message ID <CAMXFM3tU9f9WakpX_jM47BvXwi-d78KC0NzSnGAX3j2hWfF=uA@mail.gmail.com>
Download mbox | patch
Permalink /patch/13548/
State New
Headers show

Comments

Andrew Senkevich - July 1, 2016, 11:53 a.m.
2016-07-01 1:25 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
> On Thu, 30 Jun 2016, Andrew Senkevich wrote:
>
>> 2016-06-30 16:46 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
>> > On Thu, 30 Jun 2016, Andrew Senkevich wrote:
>> >
>> >> Indeed, it can be simplified now.
>> >>
>> >> Is it  Ok with that change for trunk as well as for 2.22 and 2.23
>> >> release branches?
>> >
>> > Please send the actual patch you are proposing.
>>
>> Here is attached.
>
> This one is OK.

Thanks.
What about a followup fix for trunk which adds "linear" versions? May
be create according bugzilla bug?

Proposed patch is:

        * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New symbols added.
        * sysdeps/x86_64/fpu/Versions: New versions added.
        * sysdeps/x86/fpu/bits/math-vector.h: Added sincos/sincosf vector
        declaration with linear clause.
        * sysdeps/x86_64/fpu/test-double-libmvec-sincos.c: Added ABI test for
        vector sincos declared with linear clause.
        * sysdeps/x86_64/fpu/test-float-libmvec-sincosf.c: Likewise for sincosf.



--
WBR,
Andrew
Joseph Myers - July 1, 2016, 4:15 p.m.
On Fri, 1 Jul 2016, Andrew Senkevich wrote:

> What about a followup fix for trunk which adds "linear" versions? May
> be create according bugzilla bug?

Try posting again (with 2.25 symbol versions) once 2.24 is out.
Andrew Senkevich - July 4, 2016, 12:13 p.m.
2016-07-01 19:15 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
> On Fri, 1 Jul 2016, Andrew Senkevich wrote:
>
>> What about a followup fix for trunk which adds "linear" versions? May
>> be create according bugzilla bug?
>
> Try posting again (with 2.25 symbol versions) once 2.24 is out.

But we discussed this as fix for 2.24 earlier in June in this thread
and in bugzilla bug.

I would like this fix for 2.24.  It is only new entry points for
functions existing since 2.22.


--
WBR,
Andrew
Andrew Senkevich - July 8, 2016, 2:07 p.m.
2016-07-04 15:13 GMT+03:00 Andrew Senkevich <andrew.n.senkevich@gmail.com>:
> 2016-07-01 19:15 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
>> On Fri, 1 Jul 2016, Andrew Senkevich wrote:
>>
>>> What about a followup fix for trunk which adds "linear" versions? May
>>> be create according bugzilla bug?
>>
>> Try posting again (with 2.25 symbol versions) once 2.24 is out.
>
> But we discussed this as fix for 2.24 earlier in June in this thread
> and in bugzilla bug.
>
> I would like this fix for 2.24.  It is only new entry points for
> functions existing since 2.22.

Hi,

since Joseph is unreachable until 18 July and freeze in progress can
somebody else approve this commit?

We discussed this followup fix here -
https://sourceware.org/ml/libc-alpha/2016-06/msg01273.html and here -
https://sourceware.org/bugzilla/show_bug.cgi?id=20024#c6


--
WBR,
Andrew
Adhemerval Zanella Netto - July 8, 2016, 4:20 p.m.
On 08/07/2016 11:07, Andrew Senkevich wrote:
> 2016-07-04 15:13 GMT+03:00 Andrew Senkevich <andrew.n.senkevich@gmail.com>:
>> 2016-07-01 19:15 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
>>> On Fri, 1 Jul 2016, Andrew Senkevich wrote:
>>>
>>>> What about a followup fix for trunk which adds "linear" versions? May
>>>> be create according bugzilla bug?
>>>
>>> Try posting again (with 2.25 symbol versions) once 2.24 is out.
>>
>> But we discussed this as fix for 2.24 earlier in June in this thread
>> and in bugzilla bug.
>>
>> I would like this fix for 2.24.  It is only new entry points for
>> functions existing since 2.22.
> 
> Hi,
> 
> since Joseph is unreachable until 18 July and freeze in progress can
> somebody else approve this commit?
> 
> We discussed this followup fix here -
> https://sourceware.org/ml/libc-alpha/2016-06/msg01273.html and here -
> https://sourceware.org/bugzilla/show_bug.cgi?id=20024#c6

I would say x86 maintainer as I suggested in my previous release blocker
discussion.

Patch

diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
index 80d028a..e3e450c 100644
--- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
+++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist
@@ -47,3 +47,12 @@  GLIBC_2.22 _ZGVeN8v_log F
 GLIBC_2.22 _ZGVeN8v_sin F
 GLIBC_2.22 _ZGVeN8vv_pow F
 GLIBC_2.22 _ZGVeN8vvv_sincos F
+GLIBC_2.24 GLIBC_2.24 A
+GLIBC_2.24 _ZGVbN2vl8l8_sincos F
+GLIBC_2.24 _ZGVbN4vl4l4_sincosf F
+GLIBC_2.24 _ZGVcN4vl8l8_sincos F
+GLIBC_2.24 _ZGVcN8vl4l4_sincosf F
+GLIBC_2.24 _ZGVdN4vl8l8_sincos F
+GLIBC_2.24 _ZGVdN8vl4l4_sincosf F
+GLIBC_2.24 _ZGVeN16vl4l4_sincosf F
+GLIBC_2.24 _ZGVeN8vl8l8_sincos F
diff --git a/sysdeps/x86/fpu/bits/math-vector.h
b/sysdeps/x86/fpu/bits/math-vector.h
old mode 100644
new mode 100755
index ca43cf4..c20715b
--- a/sysdeps/x86/fpu/bits/math-vector.h
+++ b/sysdeps/x86/fpu/bits/math-vector.h
@@ -28,9 +28,11 @@ 
 # if defined _OPENMP && _OPENMP >= 201307
 /* OpenMP case.  */
 #  define __DECL_SIMD_x86_64 _Pragma ("omp declare simd notinbranch")
+#  define __DECL_SIMD_x86_64_sincos __DECL_SIMD_x86_64 _Pragma ("omp
declare simd notinbranch linear (__sinx, __cosx: 1)")
 # elif __GNUC_PREREQ (6,0)
 /* W/o OpenMP use GCC 6.* __attribute__ ((__simd__)).  */
 #  define __DECL_SIMD_x86_64 __attribute__ ((__simd__ ("notinbranch")))
+#  define __DECL_SIMD_x86_64_sincos __DECL_SIMD_x86_64
 # endif

 # ifdef __DECL_SIMD_x86_64
@@ -43,9 +45,9 @@ 
 #  undef __DECL_SIMD_sinf
 #  define __DECL_SIMD_sinf __DECL_SIMD_x86_64
 #  undef __DECL_SIMD_sincos
-#  define __DECL_SIMD_sincos __DECL_SIMD_x86_64
+#  define __DECL_SIMD_sincos __DECL_SIMD_x86_64_sincos
 #  undef __DECL_SIMD_sincosf
-#  define __DECL_SIMD_sincosf __DECL_SIMD_x86_64
+#  define __DECL_SIMD_sincosf __DECL_SIMD_x86_64_sincos
 #  undef __DECL_SIMD_log
 #  define __DECL_SIMD_log __DECL_SIMD_x86_64
 #  undef __DECL_SIMD_logf
diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions
index 0813204..02df4b5 100644
--- a/sysdeps/x86_64/fpu/Versions
+++ b/sysdeps/x86_64/fpu/Versions
@@ -13,4 +13,8 @@  libmvec {
     _ZGVbN4vv_powf; _ZGVcN8vv_powf; _ZGVdN8vv_powf; _ZGVeN16vv_powf;
     _ZGVbN4vvv_sincosf; _ZGVcN8vvv_sincosf; _ZGVdN8vvv_sincosf;
_ZGVeN16vvv_sincosf;
   }
+  GLIBC_2.24 {
+    _ZGVbN2vl8l8_sincos; _ZGVcN4vl8l8_sincos; _ZGVdN4vl8l8_sincos;
_ZGVeN8vl8l8_sincos;
+    _ZGVbN4vl4l4_sincosf; _ZGVcN8vl4l4_sincosf; _ZGVdN8vl4l4_sincosf;
_ZGVeN16vl4l4_sincosf;
+  }
 }
diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-sincos.c
b/sysdeps/x86_64/fpu/test-double-libmvec-sincos.c
index 80348a2..8fe106d 100644
--- a/sysdeps/x86_64/fpu/test-double-libmvec-sincos.c
+++ b/sysdeps/x86_64/fpu/test-double-libmvec-sincos.c
@@ -21,6 +21,7 @@ 

 #define N 1000
 double x[N], s[N], c[N];
+double x1[N], s1[N], c1[N];
 double* s_ptrs[N];
 double* c_ptrs[N];
 int arch_check = 1;
@@ -28,15 +29,13 @@  int arch_check = 1;
 static void
 init_arg (void)
 {
-  int i;
-
   CHECK_ARCH_EXT;

   arch_check = 0;

-  for(i = 0; i < N; i++)
+  for(int i = 0; i < N; i++)
   {
-    x[i] = i / 3;
+    x[i] = x1[i] = i / 3;
     s_ptrs[i] = &s[i];
     c_ptrs[i] = &c[i];
   }
@@ -45,16 +44,19 @@  init_arg (void)
 static int
 test_sincos_abi (void)
 {
-  int i;
-
-  init_arg ();
+#pragma omp simd
+  for(int i = 0; i < N; i++)
+    sincos (x[i], s_ptrs[i], c_ptrs[i]);

-  if (arch_check)
-    return 77;
+  return 0;
+}

+static int
+test_sincos_linear_abi (void)
+{
 #pragma omp simd
-  for(i = 0; i < N; i++)
-    sincos (x[i], s_ptrs[i], c_ptrs[i]);
+  for(int i = 0; i < N; i++)
+    sincos (x1[i], &s1[i], &c1[i]);

   return 0;
 }
@@ -62,7 +64,16 @@  test_sincos_abi (void)
 static int
 do_test (void)
 {
-    return test_sincos_abi ();
+  init_arg ();
+
+  if (arch_check)
+    return 77;
+
+  test_sincos_abi ();
+
+  test_sincos_linear_abi ();
+
+  return 0;
 }

 #define TEST_FUNCTION do_test ()
diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-sincosf.c
b/sysdeps/x86_64/fpu/test-float-libmvec-sincosf.c
index 3b7aad8..fac2152 100644
--- a/sysdeps/x86_64/fpu/test-float-libmvec-sincosf.c
+++ b/sysdeps/x86_64/fpu/test-float-libmvec-sincosf.c
@@ -21,22 +21,21 @@ 

 #define N 1000
 float x[N], s[N], c[N];
-float *s_ptrs[N];
-float *c_ptrs[N];
+float x1[N], s1[N], c1[N];
+float* s_ptrs[N];
+float* c_ptrs[N];
 int arch_check = 1;

 static void
 init_arg (void)
 {
-  int i;
-
   CHECK_ARCH_EXT;

   arch_check = 0;

-  for(i = 0; i < N; i++)
+  for(int i = 0; i < N; i++)
   {
-    x[i] = i / 3;
+    x[i] = x1[i] = i / 3;
     s_ptrs[i] = &s[i];
     c_ptrs[i] = &c[i];
   }
@@ -45,16 +44,19 @@  init_arg (void)
 static int
 test_sincosf_abi (void)
 {
-  int i;
-
-  init_arg ();
+#pragma omp simd
+  for(int i = 0; i < N; i++)
+    sincosf (x[i], s_ptrs[i], c_ptrs[i]);

-  if (arch_check)
-    return 77;
+  return 0;
+}

+static int
+test_sincosf_linear_abi (void)
+{
 #pragma omp simd
-  for(i = 0; i < N; i++)
-    sincosf (x[i], s_ptrs[i], c_ptrs[i]);
+  for(int i = 0; i < N; i++)
+    sincosf (x1[i], &s1[i], &c1[i]);

   return 0;
 }
@@ -62,7 +64,16 @@  test_sincosf_abi (void)
 static int
 do_test (void)
 {
-  return test_sincosf_abi ();
+  init_arg ();
+
+  if (arch_check)
+    return 77;
+
+  test_sincosf_abi ();
+
+  test_sincosf_linear_abi ();
+
+  return 0;
 }

 #define TEST_FUNCTION do_test ()