about summary refs log tree commit diff
path: root/sysdeps/powerpc/fpu
diff options
context:
space:
mode:
authorAdhemerval Zanella <azanella@linux.vnet.ibm.com>2013-11-25 06:12:30 -0600
committerAdhemerval Zanella <azanella@linux.vnet.ibm.com>2013-11-25 06:34:41 -0600
commit41e8926aa4b7f17bc95984737ee82a254ad0911c (patch)
treef113bbedd0ec991ae1b9dddf3a36077e22d19815 /sysdeps/powerpc/fpu
parent085b5ddfe301eb74bc9f824ff293f054c1c910b7 (diff)
downloadglibc-41e8926aa4b7f17bc95984737ee82a254ad0911c.tar.gz
glibc-41e8926aa4b7f17bc95984737ee82a254ad0911c.tar.xz
glibc-41e8926aa4b7f17bc95984737ee82a254ad0911c.zip
PowerPC: Set/restore rounding mode only when needed
This patch helps some math functions performance by adding the libc_fexxx
variant of inline functions to handle both FPU round and exception set/restore
and by using them on the libc_fexxx_ctx functions. It is based on already coded
fexxx family functions for PPC with fpu.

Here is the summary of performance improvements due this patch (measured on a
POWER7 machine):

Before:

cos(): ITERS:9.5895e+07: TOTAL:5116.03Mcy, MAX:77.6cy, MIN:49.792cy, 18744 calls/Mcy
exp(): ITERS:2.827e+07: TOTAL:5187.15Mcy, MAX:494.018cy, MIN:38.422cy, 5450.01 calls/Mcy
pow(): ITERS:6.1705e+07: TOTAL:5144.26Mcy, MAX:171.95cy, MIN:29.935cy, 11994.9 calls/Mcy
sin(): ITERS:8.6898e+07: TOTAL:5117.06Mcy, MAX:83.841cy, MIN:46.582cy, 16982 calls/Mcy
tan(): ITERS:2.9473e+07: TOTAL:5115.39Mcy, MAX:191.017cy, MIN:172.352cy, 5761.63 calls/Mcy

After:

cos(): ITERS:2.05265e+08: TOTAL:5111.37Mcy, MAX:78.754cy, MIN:24.196cy, 40158.5 calls/Mcy
exp(): ITERS:3.341e+07: TOTAL:5170.84Mcy, MAX:476.317cy, MIN:15.574cy, 6461.23 calls/Mcy
pow(): ITERS:7.6153e+07: TOTAL:5129.1Mcy, MAX:147.5cy, MIN:30.916cy, 14847.2 calls/Mcy
sin(): ITERS:1.58816e+08: TOTAL:5115.11Mcy, MAX:1490.39cy, MIN:22.341cy, 31048.4 calls/Mcy
tan(): ITERS:3.4964e+07: TOTAL:5114.18Mcy, MAX:177.422cy, MIN:146.115cy, 6836.68 calls/Mcy
Diffstat (limited to 'sysdeps/powerpc/fpu')
-rw-r--r--sysdeps/powerpc/fpu/fenv_libc.h6
-rw-r--r--sysdeps/powerpc/fpu/fenv_private.h274
-rw-r--r--sysdeps/powerpc/fpu/math_private.h1
3 files changed, 277 insertions, 4 deletions
diff --git a/sysdeps/powerpc/fpu/fenv_libc.h b/sysdeps/powerpc/fpu/fenv_libc.h
index cb15c1cecb..ecd6b9192a 100644
--- a/sysdeps/powerpc/fpu/fenv_libc.h
+++ b/sysdeps/powerpc/fpu/fenv_libc.h
@@ -23,9 +23,9 @@
 #include <ldsodefs.h>
 #include <sysdep.h>
 
-extern const fenv_t *__fe_nomask_env (void);
+extern const fenv_t *__fe_nomask_env (void) attribute_hidden;
 
-extern const fenv_t *__fe_mask_env (void);
+extern const fenv_t *__fe_mask_env (void) attribute_hidden;
 
 /* The sticky bits in the FPSCR indicating exceptions have occurred.  */
 #define FPSCR_STICKY_BITS ((FE_ALL_EXCEPT | FE_ALL_INVALID) & ~FE_INVALID)
@@ -83,7 +83,6 @@ __fegetround (void)
 		"mfcr  %0" : "=r"(result) : : "cr7");
   return result & 3;
 }
-#define fegetround() __fegetround()
 
 static inline int
 __fesetround (int round)
@@ -107,7 +106,6 @@ __fesetround (int round)
 
   return 0;
 }
-#define fesetround(mode) __fesetround(mode)
 
 /* Definitions of all the FPSCR bit numbers */
 enum {
diff --git a/sysdeps/powerpc/fpu/fenv_private.h b/sysdeps/powerpc/fpu/fenv_private.h
new file mode 100644
index 0000000000..293f840892
--- /dev/null
+++ b/sysdeps/powerpc/fpu/fenv_private.h
@@ -0,0 +1,274 @@
+/* Private floating point rounding and exceptions handling. PowerPC version.
+   Copyright (C) 2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef FENV_PRIVATE_H
+#define FENV_PRIVATE_H 1
+
+#include <fenv.h>
+#include <fenv_libc.h>
+#include <fpu_control.h>
+
+#define _FPU_MASK_ALL (_FPU_MASK_ZM | _FPU_MASK_OM | _FPU_MASK_UM \
+                      | _FPU_MASK_XM | _FPU_MASK_IM)
+
+/* Mask everything but the rounding moded and non-IEEE arithmetic flags.  */
+#define _FPU_MASK_ROUNDING 0xffffffff00000007LL
+
+/* Mask restore rounding mode and exception enabled.  */
+#define _FPU_MASK_EXCEPT_ROUND 0xffffffff1fffff00LL
+
+/* Mask exception enable but fraction rounded/inexact and FP result/CC
+   bits.  */
+#define _FPU_MASK_FRAC_INEX_RET_CC 0x1ff80fff
+
+static __always_inline void
+libc_feholdexcept_ppc (fenv_t *envp)
+{
+  fenv_union_t old, new;
+
+  old.fenv = *envp = fegetenv_register ();
+
+  new.l = old.l & _FPU_MASK_ROUNDING;
+
+  /* If the old env had any enabled exceptions, then mask SIGFPE in the
+     MSR FE0/FE1 bits.  This may allow the FPU to run faster because it
+     always takes the default action and can not generate SIGFPE.  */
+  if ((old.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_mask_env ();
+
+  fesetenv_register (new.fenv);
+}
+
+static __always_inline void
+libc_fesetround_ppc (int r)
+{
+  __fesetround (r);
+}
+
+static __always_inline void
+libc_feholdexcept_setround_ppc (fenv_t *envp, int r)
+{
+  fenv_union_t old, new;
+
+  old.fenv = *envp = fegetenv_register ();
+
+  new.l = (old.l & _FPU_MASK_ROUNDING) | r;
+
+  if ((old.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_mask_env ();
+
+  fesetenv_register (new.fenv);
+}
+
+static __always_inline int
+libc_fetestexcept_ppc (int e)
+{
+  fenv_union_t u;
+  u.fenv = fegetenv_register ();
+  return u.l & e;
+}
+
+static __always_inline void
+libc_fesetenv_ppc (const fenv_t *envp)
+{
+  fenv_union_t old, new;
+
+  new.fenv = *envp;
+  old.fenv = fegetenv_register ();
+
+  /* If the old env has no enabled exceptions and the new env has any enabled
+     exceptions, then unmask SIGFPE in the MSR FE0/FE1 bits.  This will put the
+     hardware into "precise mode" and may cause the FPU to run slower on some
+     hardware.  */
+  if ((old.l & _FPU_MASK_ALL) == 0 && (new.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_nomask_env ();
+
+  /* If the old env had any enabled exceptions and the new env has no enabled
+     exceptions, then mask SIGFPE in the MSR FE0/FE1 bits.  This may allow the
+     FPU to run faster because it always takes the default action and can not
+     generate SIGFPE.  */
+  if ((old.l & _FPU_MASK_ALL) != 0 && (new.l & _FPU_MASK_ALL) == 0)
+    (void) __fe_mask_env ();
+
+  fesetenv_register (*envp);
+}
+
+static __always_inline int
+libc_feupdateenv_test_ppc (fenv_t *envp, int ex)
+{
+  fenv_union_t old, new;
+
+  new.fenv = *envp;
+  old.fenv = fegetenv_register ();
+
+  /* Restore rounding mode and exception enable from *envp and merge
+     exceptions.  Leave fraction rounded/inexact and FP result/CC bits
+     unchanged.  */
+  new.l = (old.l & _FPU_MASK_EXCEPT_ROUND)
+          | (new.l & _FPU_MASK_FRAC_INEX_RET_CC);
+
+  if ((old.l & _FPU_MASK_ALL) == 0 && (new.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_nomask_env ();
+
+  if ((old.l & _FPU_MASK_ALL) != 0 && (new.l & _FPU_MASK_ALL) == 0)
+    (void) __fe_mask_env ();
+
+  fesetenv_register (new.fenv);
+
+  return old.l & ex;
+}
+
+static __always_inline void
+libc_feupdateenv_ppc (fenv_t *e)
+{
+  libc_feupdateenv_test_ppc (e, 0);
+}
+
+static __always_inline void
+libc_feholdsetround_ppc (fenv_t *e, int r)
+{
+  fenv_union_t old, new;
+
+  old.fenv = fegetenv_register ();
+  /* Clear current precision and set newer one.  */
+  new.l = (old.l & ~0x3) | r;
+  *e = old.fenv;
+
+  if ((old.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_mask_env ();
+  fesetenv_register (new.fenv);
+}
+
+static __always_inline void
+libc_feresetround_ppc (fenv_t *envp)
+{
+  fenv_union_t old, new;
+
+  new.fenv = *envp;
+  old.fenv = fegetenv_register ();
+
+  /* Restore rounding mode and exception enable from *envp and merge
+     exceptions.  Leave fraction rounded/inexact and FP result/CC bits
+     unchanged.  */
+  new.l = (old.l & _FPU_MASK_EXCEPT_ROUND)
+          | (new.l & _FPU_MASK_FRAC_INEX_RET_CC);
+
+  if ((old.l & _FPU_MASK_ALL) == 0 && (new.l & _FPU_MASK_ALL) != 0)
+    (void) __fe_nomask_env ();
+
+  if ((old.l & _FPU_MASK_ALL) != 0 && (new.l & _FPU_MASK_ALL) == 0)
+    (void) __fe_mask_env ();
+
+  /* Atomically enable and raise (if appropriate) exceptions set in `new'.  */
+  fesetenv_register (new.fenv);
+}
+
+#define libc_feholdexceptf           libc_feholdexcept_ppc
+#define libc_feholdexcept            libc_feholdexcept_ppc
+#define libc_feholdexcept_setroundf  libc_feholdexcept_setround_ppc
+#define libc_feholdexcept_setround   libc_feholdexcept_setround_ppc
+#define libc_fetestexceptf           libc_fetestexcept_ppc
+#define libc_fetestexcept            libc_fetestexcept_ppc
+#define libc_fesetroundf             libc_fesetround_ppc
+#define libc_fesetround              libc_fesetround_ppc
+#define libc_fesetenvf               libc_fesetenv_ppc
+#define libc_fesetenv                libc_fesetenv_ppc
+#define libc_feupdateenv_testf       libc_feupdateenv_test_ppc
+#define libc_feupdateenv_test        libc_feupdateenv_test_ppc
+#define libc_feupdateenvf            libc_feupdateenv_ppc
+#define libc_feupdateenv             libc_feupdateenv_ppc
+#define libc_feholdsetroundf         libc_feholdsetround_ppc
+#define libc_feholdsetround          libc_feholdsetround_ppc
+#define libc_feresetroundf           libc_feresetround_ppc
+#define libc_feresetround            libc_feresetround_ppc
+
+
+/* We have support for rounding mode context.  */
+#define HAVE_RM_CTX 1
+
+static __always_inline void
+libc_feholdexcept_setround_ppc_ctx (struct rm_ctx *ctx, int r)
+{
+  fenv_union_t old, new;
+
+  old.fenv = fegetenv_register ();
+
+  new.l = (old.l & _FPU_MASK_ROUNDING) | r;
+  ctx->env = old.fenv;
+  if (__glibc_unlikely (new.l != old.l))
+    {
+      if ((old.l & _FPU_MASK_ALL) != 0)
+	(void) __fe_mask_env ();
+      fesetenv_register (new.fenv);
+      ctx->updated_status = true;
+    }
+  else
+    ctx->updated_status = false;
+}
+
+static __always_inline void
+libc_fesetenv_ppc_ctx (struct rm_ctx *ctx)
+{
+  libc_fesetenv_ppc (&ctx->env);
+}
+
+static __always_inline void
+libc_feupdateenv_ppc_ctx (struct rm_ctx *ctx)
+{
+  if (__glibc_unlikely (ctx->updated_status))
+    libc_feupdateenv_test_ppc (&ctx->env, 0);
+}
+
+static __always_inline void
+libc_feholdsetround_ppc_ctx (struct rm_ctx *ctx, int r)
+{
+  fenv_union_t old, new;
+
+  old.fenv = fegetenv_register ();
+  new.l = (old.l & ~0x3) | r;
+  ctx->env = old.fenv;
+  if (__glibc_unlikely (new.l != old.l))
+    {
+      if ((old.l & _FPU_MASK_ALL) != 0)
+	(void) __fe_mask_env ();
+      fesetenv_register (new.fenv);
+      ctx->updated_status = true;
+    }
+  else
+    ctx->updated_status = false;
+}
+
+static __always_inline void
+libc_feresetround_ppc_ctx (struct rm_ctx *ctx)
+{
+  if (__glibc_unlikely (ctx->updated_status))
+    libc_feresetround_ppc (&ctx->env);
+}
+
+#define libc_feholdexcept_setroundf_ctx  libc_feholdexcept_setround_ppc_ctx
+#define libc_feholdexcept_setround_ctx   libc_feholdexcept_setround_ppc_ctx
+#define libc_fesetenv_ctx                libc_fesetenv_ppc_ctx
+#define libc_fesetenvf_ctx               libc_fesetenv_ppc_ctx
+#define libc_feholdsetround_ctx          libc_feholdsetround_ppc_ctx
+#define libc_feholdsetroundf_ctx         libc_feholdsetround_ppc_ctx
+#define libc_feresetround_ctx            libc_feresetround_ppc_ctx
+#define libc_feresetroundf_ctx           libc_feresetround_ppc_ctx
+#define libc_feupdateenvf_ctx            libc_feupdateenv_ppc_ctx
+#define libc_feupdateenv_ctx             libc_feupdateenv_ppc_ctx
+
+#endif
diff --git a/sysdeps/powerpc/fpu/math_private.h b/sysdeps/powerpc/fpu/math_private.h
index 6c00785633..c8833d694f 100644
--- a/sysdeps/powerpc/fpu/math_private.h
+++ b/sysdeps/powerpc/fpu/math_private.h
@@ -22,6 +22,7 @@
 #include <sysdep.h>
 #include <ldsodefs.h>
 #include <dl-procinfo.h>
+#include <fenv_private.h>
 #include_next <math_private.h>
 
 # if __WORDSIZE == 64 || defined _ARCH_PWR4