[v2] libstdc++: Add support for POWER9 DARN instruction to std::random_device

Message ID YXAS3kwGRLUDmTR5@redhat.com
State Committed
Headers
Series [v2] libstdc++: Add support for POWER9 DARN instruction to std::random_device |

Commit Message

Jonathan Wakely Oct. 20, 2021, 1 p.m. UTC
  On 20/10/21 10:12 +0100, Jonathan Wakely wrote:
>On 19/10/21 17:47 +0100, Jonathan Wakely wrote:
>>The ISA-3.0 instruction set includes DARN ("deliver a random number")
>>which can be used similar to the existing support for RDRAND and RDSEED.
>>
>>libstdc++-v3/ChangeLog:
>>
>>	* src/c++11/random.cc (USE_DARN): Define.
>>	(__ppc_darn): New function to use POWER9 DARN instruction.
>>	(Which): Add 'darn' enumerator.
>>	(which_source): Check for __ppc_darn.
>>	(random_device::_M_init): Support "darn" and "hw" tokens.
>>	(random_device::_M_getentropy): Add darn to switch.
>>	* testsuite/26_numerics/random/random_device/cons/token.cc:
>>	Check "darn" token.
>>	* testsuite/26_numerics/random/random_device/entropy.cc:
>>	Likewise.
>>
>>Tested powerpc64le-linux (power8 and power9) and x86_64-linux.
>>
>>The new "darn" (power-specific) and "hw" (x86 and power)
>>strings should be documented, but I'll do that if this gets committed.
>>
>>Most of this patch is just "more of the same", similar to the existing
>>code for RDRAND and RDSEED on x86, but the parts of the patch I'd like
>>more eyes on are:
>>
>>
>>+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__
>>+# define USE_DARN 1
>>#endif
>
>This means DARN can only be used when __builtin_cpu_supports is
>available, which means glibc 2.23 ... is that acceptable? It means
>RHEL 7 wouldn't be able to use DARN, but RHEL 8 would.
>
>There certainly are POWER9 machines running RHEL 7 and similar
>vintages (the GCC compile farm has one) so if there's another way to
>check for ISA 3.0 then I could use that.
>
>If __POWER9_VECTOR__ is defined when building libstdc++, presumably
>that means the whole library can only be run on POWER9 hardware. So
>would that mean we don't need to check __builtin_cpu_supports("darn")
>when __POWER9_VECTOR__ is defined? Or is it possible to build with
>-mcpu=power8 -mpower9-vector and run it on h/w without the DARN
>instruction?
>
>Also, I forgot to add a configure check that the assembler supports
>darn, which is another prerequisite for using it here.
>
>>@@ -135,6 +137,15 @@ namespace std _GLIBCXX_VISIBILITY(default)
>>#endif
>>#endif
>>
>>+#ifdef USE_DARN
>>+    unsigned int
>>+    __attribute__((target("power9")))
>
>Oops, that should be "cpu=power9".
>
>With that change it works on a POWER9 machine (9009-42A) with glibc
>2.34 and binutils 2.35.
>

Here's the updated patch with a configure check for assembler support,
and the target attribute fixed.

This still requires Glibc 2.23 for __builtin_cpu_supports, which I'm
assuming is acceptable.
  

Comments

Jonathan Wakely Nov. 3, 2021, 3:01 p.m. UTC | #1
Any feedback from POWER maintainers about this? I'd like to push it soon if
there's nothing wrong with it.

On Wed, 20 Oct 2021 at 14:00, Jonathan Wakely via Libstdc++ <
libstdc++@gcc.gnu.org> wrote:

> On 20/10/21 10:12 +0100, Jonathan Wakely wrote:
> >On 19/10/21 17:47 +0100, Jonathan Wakely wrote:
> >>The ISA-3.0 instruction set includes DARN ("deliver a random number")
> >>which can be used similar to the existing support for RDRAND and RDSEED.
> >>
> >>libstdc++-v3/ChangeLog:
> >>
> >>      * src/c++11/random.cc (USE_DARN): Define.
> >>      (__ppc_darn): New function to use POWER9 DARN instruction.
> >>      (Which): Add 'darn' enumerator.
> >>      (which_source): Check for __ppc_darn.
> >>      (random_device::_M_init): Support "darn" and "hw" tokens.
> >>      (random_device::_M_getentropy): Add darn to switch.
> >>      * testsuite/26_numerics/random/random_device/cons/token.cc:
> >>      Check "darn" token.
> >>      * testsuite/26_numerics/random/random_device/entropy.cc:
> >>      Likewise.
> >>
> >>Tested powerpc64le-linux (power8 and power9) and x86_64-linux.
> >>
> >>The new "darn" (power-specific) and "hw" (x86 and power)
> >>strings should be documented, but I'll do that if this gets committed.
> >>
> >>Most of this patch is just "more of the same", similar to the existing
> >>code for RDRAND and RDSEED on x86, but the parts of the patch I'd like
> >>more eyes on are:
> >>
> >>
> >>+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__
> >>+# define USE_DARN 1
> >>#endif
> >
> >This means DARN can only be used when __builtin_cpu_supports is
> >available, which means glibc 2.23 ... is that acceptable? It means
> >RHEL 7 wouldn't be able to use DARN, but RHEL 8 would.
> >
> >There certainly are POWER9 machines running RHEL 7 and similar
> >vintages (the GCC compile farm has one) so if there's another way to
> >check for ISA 3.0 then I could use that.
> >
> >If __POWER9_VECTOR__ is defined when building libstdc++, presumably
> >that means the whole library can only be run on POWER9 hardware. So
> >would that mean we don't need to check __builtin_cpu_supports("darn")
> >when __POWER9_VECTOR__ is defined? Or is it possible to build with
> >-mcpu=power8 -mpower9-vector and run it on h/w without the DARN
> >instruction?
> >
> >Also, I forgot to add a configure check that the assembler supports
> >darn, which is another prerequisite for using it here.
> >
> >>@@ -135,6 +137,15 @@ namespace std _GLIBCXX_VISIBILITY(default)
> >>#endif
> >>#endif
> >>
> >>+#ifdef USE_DARN
> >>+    unsigned int
> >>+    __attribute__((target("power9")))
> >
> >Oops, that should be "cpu=power9".
> >
> >With that change it works on a POWER9 machine (9009-42A) with glibc
> >2.34 and binutils 2.35.
> >
>
> Here's the updated patch with a configure check for assembler support,
> and the target attribute fixed.
>
> This still requires Glibc 2.23 for __builtin_cpu_supports, which I'm
> assuming is acceptable.
>
>
>
  
Jonathan Wakely Nov. 3, 2021, 3:02 p.m. UTC | #2
On Wed, 3 Nov 2021 at 15:01, Jonathan Wakely wrote:

> Any feedback from POWER maintainers about this? I'd like to push it soon
> if there's nothing wrong with it.
>

With the updated patch attached again this time ...



> On Wed, 20 Oct 2021 at 14:00, Jonathan Wakely via Libstdc++ <
> libstdc++@gcc.gnu.org> wrote:
>
>> On 20/10/21 10:12 +0100, Jonathan Wakely wrote:
>> >On 19/10/21 17:47 +0100, Jonathan Wakely wrote:
>> >>The ISA-3.0 instruction set includes DARN ("deliver a random number")
>> >>which can be used similar to the existing support for RDRAND and RDSEED.
>> >>
>> >>libstdc++-v3/ChangeLog:
>> >>
>> >>      * src/c++11/random.cc (USE_DARN): Define.
>> >>      (__ppc_darn): New function to use POWER9 DARN instruction.
>> >>      (Which): Add 'darn' enumerator.
>> >>      (which_source): Check for __ppc_darn.
>> >>      (random_device::_M_init): Support "darn" and "hw" tokens.
>> >>      (random_device::_M_getentropy): Add darn to switch.
>> >>      * testsuite/26_numerics/random/random_device/cons/token.cc:
>> >>      Check "darn" token.
>> >>      * testsuite/26_numerics/random/random_device/entropy.cc:
>> >>      Likewise.
>> >>
>> >>Tested powerpc64le-linux (power8 and power9) and x86_64-linux.
>> >>
>> >>The new "darn" (power-specific) and "hw" (x86 and power)
>> >>strings should be documented, but I'll do that if this gets committed.
>> >>
>> >>Most of this patch is just "more of the same", similar to the existing
>> >>code for RDRAND and RDSEED on x86, but the parts of the patch I'd like
>> >>more eyes on are:
>> >>
>> >>
>> >>+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__
>> >>+# define USE_DARN 1
>> >>#endif
>> >
>> >This means DARN can only be used when __builtin_cpu_supports is
>> >available, which means glibc 2.23 ... is that acceptable? It means
>> >RHEL 7 wouldn't be able to use DARN, but RHEL 8 would.
>> >
>> >There certainly are POWER9 machines running RHEL 7 and similar
>> >vintages (the GCC compile farm has one) so if there's another way to
>> >check for ISA 3.0 then I could use that.
>> >
>> >If __POWER9_VECTOR__ is defined when building libstdc++, presumably
>> >that means the whole library can only be run on POWER9 hardware. So
>> >would that mean we don't need to check __builtin_cpu_supports("darn")
>> >when __POWER9_VECTOR__ is defined? Or is it possible to build with
>> >-mcpu=power8 -mpower9-vector and run it on h/w without the DARN
>> >instruction?
>> >
>> >Also, I forgot to add a configure check that the assembler supports
>> >darn, which is another prerequisite for using it here.
>> >
>> >>@@ -135,6 +137,15 @@ namespace std _GLIBCXX_VISIBILITY(default)
>> >>#endif
>> >>#endif
>> >>
>> >>+#ifdef USE_DARN
>> >>+    unsigned int
>> >>+    __attribute__((target("power9")))
>> >
>> >Oops, that should be "cpu=power9".
>> >
>> >With that change it works on a POWER9 machine (9009-42A) with glibc
>> >2.34 and binutils 2.35.
>> >
>>
>> Here's the updated patch with a configure check for assembler support,
>> and the target attribute fixed.
>>
>> This still requires Glibc 2.23 for __builtin_cpu_supports, which I'm
>> assuming is acceptable.
>>
>>
>>
commit a6f925407dd05c593b230da1627435adc53584f8
Author: Jonathan Wakely <jwakely@redhat.com>
Date:   Wed Oct 20 09:25:24 2021

    libstdc++: Add support for POWER9 DARN instruction to std::random_device
    
    The ISA-3.0 instruction set includes DARN ("deliver a random number")
    which can be used similar to the existing support for RDRAND and RDSEED.
    
    libstdc++-v3/ChangeLog:
    
            * acinclude.m4 (GLIBCXX_CHECK_PPC_DARN): Check assembler.
            * config.h.in: Regenerate.
            * configure: Regenerate.
            * configure.ac: Use GLIBCXX_CHECK_PPC_DARN.
            * src/c++11/random.cc [_GLIBCXX_PPC_DARN] (USE_DARN): Define.
            (__ppc_darn): New function to use POWER9 DARN instruction.
            (Which): Add 'darn' enumerator.
            (which_source): Check for __ppc_darn.
            (random_device::_M_init): Support "darn" and "hw" tokens.
            (random_device::_M_getentropy): Add darn to switch.
            * testsuite/26_numerics/random/random_device/cons/token.cc:
            Check "darn" token.
            * testsuite/26_numerics/random/random_device/entropy.cc:
            Likewise.

diff --git a/libstdc++-v3/acinclude.m4 b/libstdc++-v3/acinclude.m4
index 90ecc4a87a2..9ff9ceb20ac 100644
--- a/libstdc++-v3/acinclude.m4
+++ b/libstdc++-v3/acinclude.m4
@@ -4100,6 +4100,27 @@ AC_DEFUN([GLIBCXX_CHECK_X86_RDSEED], [
   AC_MSG_RESULT($ac_cv_x86_rdseed)
 ])
 
+dnl
+dnl Check whether darn is supported in the assembler.
+AC_DEFUN([GLIBCXX_CHECK_PPC_DARN], [
+  AC_MSG_CHECKING([for darn support in assembler])
+  AC_CACHE_VAL(ac_cv_ppc_darn, [
+  ac_cv_ppc_darn=no
+  case "$target" in
+    powerpc*-*-*)
+    AC_TRY_COMPILE(, [
+      signed int x;
+      __asm__ __volatile__ (".machine power9; darn %0,0;": "=r" (x));
+    ], [ac_cv_ppc_darn=yes], [ac_cv_ppc_darn=no])
+  esac
+  ])
+  if test $ac_cv_ppc_darn = yes; then
+    AC_DEFINE(_GLIBCXX_PPC_DARN, 1,
+		[ Defined if as can handle darn. ])
+  fi
+  AC_MSG_RESULT($ac_cv_ppc_darn)
+])
+
 dnl
 dnl Check whether get_nprocs is available in <sys/sysinfo.h>, and define _GLIBCXX_USE_GET_NPROCS.
 dnl
diff --git a/libstdc++-v3/configure.ac b/libstdc++-v3/configure.ac
index 2d68b3672b9..1189c68c380 100644
--- a/libstdc++-v3/configure.ac
+++ b/libstdc++-v3/configure.ac
@@ -467,6 +467,8 @@ GCC_CHECK_ASSEMBLER_HWCAP
 GLIBCXX_CHECK_X86_RDRAND
 # Check if assembler supports rdseed opcode.
 GLIBCXX_CHECK_X86_RDSEED
+# Check if assembler supports darn opcode.
+GLIBCXX_CHECK_PPC_DARN
 
 # This depends on GLIBCXX_ENABLE_SYMVERS and GLIBCXX_IS_NATIVE.
 GLIBCXX_CONFIGURE_TESTSUITE
diff --git a/libstdc++-v3/src/c++11/random.cc b/libstdc++-v3/src/c++11/random.cc
index 4b64bde00ea..213ad691837 100644
--- a/libstdc++-v3/src/c++11/random.cc
+++ b/libstdc++-v3/src/c++11/random.cc
@@ -37,6 +37,9 @@
 # ifdef _GLIBCXX_X86_RDSEED
 #  define USE_RDSEED 1
 # endif
+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__ \
+  && defined _GLIBCXX_PPC_DARN
+# define USE_DARN 1
 #endif
 
 #include <cerrno>
@@ -69,7 +72,7 @@
 #if defined _GLIBCXX_USE_CRT_RAND_S || defined _GLIBCXX_USE_DEV_RANDOM
 // The OS provides a source of randomness we can use.
 # pragma GCC poison _M_mt
-#elif defined USE_RDRAND || defined USE_RDSEED
+#elif defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
 // Hardware instructions might be available, but use cpuid checks at runtime.
 # pragma GCC poison _M_mt
 // If the runtime cpuid checks fail we'll use a linear congruential engine.
@@ -135,6 +138,15 @@ namespace std _GLIBCXX_VISIBILITY(default)
 #endif
 #endif
 
+#ifdef USE_DARN
+    unsigned int
+    __attribute__((target("cpu=power9")))
+    __ppc_darn(void*)
+    {
+      return __builtin_darn_32();
+    }
+#endif
+
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     unsigned int
     __winxp_rand_s(void*)
@@ -193,11 +205,16 @@ namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif
 
-    enum Which {
-      rand_s = 1, rdseed = 2, rdrand = 4, device_file = 8, prng = 16,
+    enum Which : unsigned {
+      device_file = 1, prng = 2, rand_s = 4,
+      rdseed = 64, rdrand = 128, darn = 256,
       any = 0xffff
     };
 
+    constexpr Which
+    operator|(Which l, Which r) noexcept
+    { return Which(unsigned(l) | unsigned(r)); }
+
     inline Which
     which_source(random_device::result_type (*func [[maybe_unused]])(void*),
 		 void* file [[maybe_unused]])
@@ -221,6 +238,11 @@ namespace std _GLIBCXX_VISIBILITY(default)
 	return rdrand;
 #endif
 
+#ifdef USE_DARN
+      if (func == &__ppc_darn)
+	return darn;
+#endif
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
       if (file != nullptr)
 	return device_file;
@@ -269,6 +291,14 @@ namespace std _GLIBCXX_VISIBILITY(default)
     else if (token == "rdrand" || token == "rdrnd")
       which = rdrand;
 #endif // USE_RDRAND
+#ifdef USE_DARN
+    else if (token == "darn")
+      which = darn;
+#endif
+#if defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
+    else if (token == "hw" || token == "hardware")
+      which = rdrand | rdseed | darn;
+#endif
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     else if (token == "rand_s")
       which = rand_s;
@@ -346,6 +376,17 @@ namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif // USE_RDRAND
 
+#ifdef USE_DARN
+    if (which & darn)
+      {
+	if (__builtin_cpu_supports("darn"))
+	  {
+	    _M_func = &__ppc_darn;
+	    return;
+	  }
+      }
+#endif // USE_DARN
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
     if (which & device_file)
     {
@@ -497,6 +538,7 @@ namespace std _GLIBCXX_VISIBILITY(default)
     {
     case rdrand:
     case rdseed:
+    case darn:
       return (double) max;
     case rand_s:
     case prng:
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
index aeb7403e830..d6ac3a37c64 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
@@ -51,8 +51,9 @@ test03()
 {
   // At least one of these tokens should be valid.
   const std::string tokens[] = {
-    "rdseed", "rdrand", "rand_s", "/dev/urandom", "/dev/random", "mt19937",
-    "prng"
+    "rdseed", "rdrand", "darn",
+    "rand_s", "/dev/urandom", "/dev/random",
+    "mt19937", "prng"
   };
   int count = 0;
   for (const std::string& token : tokens)
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
index 9ef1538d2bb..6f3ebb1b38e 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
@@ -22,7 +22,7 @@ test01()
       VERIFY( entropy <= max );
     }
 
-  for (auto token : { "rdrand", "rdseed" })
+  for (auto token : { "rdrand", "rdseed", "darn", "hw" })
     if (__gnu_test::random_device_available(token))
     {
       const double entropy = std::random_device(token).entropy();
  
Li, Pan2 via Gcc-patches Nov. 4, 2021, 8:44 p.m. UTC | #3
For posterity:  This was discussed briefly on IRC, and Segher approved with some 
simplifications and a request to implement a fail/retry check.

Thanks,
Bill

On 11/3/21 10:02 AM, Jonathan Wakely wrote:
> On Wed, 3 Nov 2021 at 15:01, Jonathan Wakely wrote: Any feedback from POWER maintainers about this? I'd like to push it soon if there's nothing wrong with it. With the updated patch attached again this time ... ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ZjQcmQRYFpfptBannerStart
> This Message Is From an External Sender
> This message came from outside your organization.
> ZjQcmQRYFpfptBannerEnd
> On Wed, 3 Nov 2021 at 15:01, Jonathan Wakely wrote:
>
>     Any feedback from POWER maintainers about this? I'd like to push it soon if there's nothing wrong with it.
>
>
> With the updated patch attached again this time ...
>
>
>
>     On Wed, 20 Oct 2021 at 14:00, Jonathan Wakely via Libstdc++ <libstdc++@gcc.gnu.org <mailto:libstdc%2B%2B@gcc.gnu.org>> wrote:
>
>         On 20/10/21 10:12 +0100, Jonathan Wakely wrote:
>         >On 19/10/21 17:47 +0100, Jonathan Wakely wrote:
>         >>The ISA-3.0 instruction set includes DARN ("deliver a random number")
>         >>which can be used similar to the existing support for RDRAND and RDSEED.
>         >>
>         >>libstdc++-v3/ChangeLog:
>         >>
>         >>      * src/c++11/random.cc (USE_DARN): Define.
>         >>      (__ppc_darn): New function to use POWER9 DARN instruction.
>         >>      (Which): Add 'darn' enumerator.
>         >>      (which_source): Check for __ppc_darn.
>         >>      (random_device::_M_init): Support "darn" and "hw" tokens.
>         >>      (random_device::_M_getentropy): Add darn to switch.
>         >>      * testsuite/26_numerics/random/random_device/cons/token.cc:
>         >>      Check "darn" token.
>         >>      * testsuite/26_numerics/random/random_device/entropy.cc:
>         >>      Likewise.
>         >>
>         >>Tested powerpc64le-linux (power8 and power9) and x86_64-linux.
>         >>
>         >>The new "darn" (power-specific) and "hw" (x86 and power)
>         >>strings should be documented, but I'll do that if this gets committed.
>         >>
>         >>Most of this patch is just "more of the same", similar to the existing
>         >>code for RDRAND and RDSEED on x86, but the parts of the patch I'd like
>         >>more eyes on are:
>         >>
>         >>
>         >>+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__
>         >>+# define USE_DARN 1
>         >>#endif
>         >
>         >This means DARN can only be used when __builtin_cpu_supports is
>         >available, which means glibc 2.23 ... is that acceptable? It means
>         >RHEL 7 wouldn't be able to use DARN, but RHEL 8 would.
>         >
>         >There certainly are POWER9 machines running RHEL 7 and similar
>         >vintages (the GCC compile farm has one) so if there's another way to
>         >check for ISA 3.0 then I could use that.
>         >
>         >If __POWER9_VECTOR__ is defined when building libstdc++, presumably
>         >that means the whole library can only be run on POWER9 hardware. So
>         >would that mean we don't need to check __builtin_cpu_supports("darn")
>         >when __POWER9_VECTOR__ is defined? Or is it possible to build with
>         >-mcpu=power8 -mpower9-vector and run it on h/w without the DARN
>         >instruction?
>         >
>         >Also, I forgot to add a configure check that the assembler supports
>         >darn, which is another prerequisite for using it here.
>         >
>         >>@@ -135,6 +137,15 @@ namespace std _GLIBCXX_VISIBILITY(default)
>         >>#endif
>         >>#endif
>         >>
>         >>+#ifdef USE_DARN
>         >>+    unsigned int
>         >>+    __attribute__((target("power9")))
>         >
>         >Oops, that should be "cpu=power9".
>         >
>         >With that change it works on a POWER9 machine (9009-42A) with glibc
>         >2.34 and binutils 2.35.
>         >
>
>         Here's the updated patch with a configure check for assembler support,
>         and the target attribute fixed.
>
>         This still requires Glibc 2.23 for __builtin_cpu_supports, which I'm
>         assuming is acceptable.
>
>
>
>
  
Jonathan Wakely Nov. 5, 2021, 12:44 p.m. UTC | #4
On Thu, 4 Nov 2021 at 20:44, Bill Schmidt wrote:

> For posterity:  This was discussed briefly on IRC, and Segher approved
> with some
> simplifications and a request to implement a fail/retry check.
>
>
Here's what I have now. No more assembler check in configure, and it uses
the 64-bit __builtin_darn() and truncates it to 32-bit, or retries (up to
100 times) if it fails.

I'm doing some more testing now.
commit 77816264712923481db37a29d1638faa1d99aadc
Author: Jonathan Wakely <jwakely@redhat.com>
Date:   Wed Oct 20 09:25:24 2021

    libstdc++: Add support for POWER9 DARN instruction to std::random_device
    
    The ISA-3.0 instruction set includes DARN ("deliver a random number")
    which can be used similar to the existing support for RDRAND and RDSEED.
    
    libstdc++-v3/ChangeLog:
    
            * src/c++11/random.cc [__powerpc__] (USE_DARN): Define.
            (__ppc_darn): New function to use POWER9 DARN instruction.
            (Which): Add 'darn' enumerator.
            (which_source): Check for __ppc_darn.
            (random_device::_M_init): Support "darn" and "hw" tokens.
            (random_device::_M_getentropy): Add darn to switch.
            * testsuite/26_numerics/random/random_device/cons/token.cc:
            Check "darn" token.
            * testsuite/26_numerics/random/random_device/entropy.cc:
            Likewise.

diff --git a/libstdc++-v3/src/c++11/random.cc b/libstdc++-v3/src/c++11/random.cc
index 4b64bde00ea..b0d88374d59 100644
--- a/libstdc++-v3/src/c++11/random.cc
+++ b/libstdc++-v3/src/c++11/random.cc
@@ -37,6 +37,8 @@
 # ifdef _GLIBCXX_X86_RDSEED
 #  define USE_RDSEED 1
 # endif
+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__
+# define USE_DARN 1
 #endif
 
 #include <cerrno>
@@ -69,7 +71,7 @@
 #if defined _GLIBCXX_USE_CRT_RAND_S || defined _GLIBCXX_USE_DEV_RANDOM
 // The OS provides a source of randomness we can use.
 # pragma GCC poison _M_mt
-#elif defined USE_RDRAND || defined USE_RDSEED
+#elif defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
 // Hardware instructions might be available, but use cpuid checks at runtime.
 # pragma GCC poison _M_mt
 // If the runtime cpuid checks fail we'll use a linear congruential engine.
@@ -135,6 +137,24 @@ namespace std _GLIBCXX_VISIBILITY(default)
 #endif
 #endif
 
+#ifdef USE_DARN
+    unsigned int
+    __attribute__((target("cpu=power9")))
+    __ppc_darn(void*)
+    {
+      const uint64_t failed = -1;
+      unsigned int retries = 100;
+      uint64_t val = __builtin_darn();
+      while (val == failed) [[__unlikely__]]
+	{
+	  if (--retries == 0)
+	    std::__throw_runtime_error(__N("random_device: darn failed"));
+	  val = __builtin_darn();
+	}
+      return (uint32_t)val;
+    }
+#endif
+
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     unsigned int
     __winxp_rand_s(void*)
@@ -193,11 +213,16 @@ namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif
 
-    enum Which {
-      rand_s = 1, rdseed = 2, rdrand = 4, device_file = 8, prng = 16,
+    enum Which : unsigned {
+      device_file = 1, prng = 2, rand_s = 4,
+      rdseed = 64, rdrand = 128, darn = 256,
       any = 0xffff
     };
 
+    constexpr Which
+    operator|(Which l, Which r) noexcept
+    { return Which(unsigned(l) | unsigned(r)); }
+
     inline Which
     which_source(random_device::result_type (*func [[maybe_unused]])(void*),
 		 void* file [[maybe_unused]])
@@ -221,6 +246,11 @@ namespace std _GLIBCXX_VISIBILITY(default)
 	return rdrand;
 #endif
 
+#ifdef USE_DARN
+      if (func == &__ppc_darn)
+	return darn;
+#endif
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
       if (file != nullptr)
 	return device_file;
@@ -269,6 +299,14 @@ namespace std _GLIBCXX_VISIBILITY(default)
     else if (token == "rdrand" || token == "rdrnd")
       which = rdrand;
 #endif // USE_RDRAND
+#ifdef USE_DARN
+    else if (token == "darn")
+      which = darn;
+#endif
+#if defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
+    else if (token == "hw" || token == "hardware")
+      which = rdrand | rdseed | darn;
+#endif
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     else if (token == "rand_s")
       which = rand_s;
@@ -346,6 +384,17 @@ namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif // USE_RDRAND
 
+#ifdef USE_DARN
+    if (which & darn)
+      {
+	if (__builtin_cpu_supports("darn"))
+	  {
+	    _M_func = &__ppc_darn;
+	    return;
+	  }
+      }
+#endif // USE_DARN
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
     if (which & device_file)
     {
@@ -497,6 +546,7 @@ namespace std _GLIBCXX_VISIBILITY(default)
     {
     case rdrand:
     case rdseed:
+    case darn:
       return (double) max;
     case rand_s:
     case prng:
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
index aeb7403e830..d6ac3a37c64 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
@@ -51,8 +51,9 @@ test03()
 {
   // At least one of these tokens should be valid.
   const std::string tokens[] = {
-    "rdseed", "rdrand", "rand_s", "/dev/urandom", "/dev/random", "mt19937",
-    "prng"
+    "rdseed", "rdrand", "darn",
+    "rand_s", "/dev/urandom", "/dev/random",
+    "mt19937", "prng"
   };
   int count = 0;
   for (const std::string& token : tokens)
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
index 9ef1538d2bb..6f3ebb1b38e 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
@@ -22,7 +22,7 @@ test01()
       VERIFY( entropy <= max );
     }
 
-  for (auto token : { "rdrand", "rdseed" })
+  for (auto token : { "rdrand", "rdseed", "darn", "hw" })
     if (__gnu_test::random_device_available(token))
     {
       const double entropy = std::random_device(token).entropy();
  
Li, Pan2 via Gcc-patches Nov. 5, 2021, 1:20 p.m. UTC | #5
On 11/5/21 7:44 AM, Jonathan Wakely wrote:
> On Thu, 4 Nov 2021 at 20:44, Bill Schmidt wrote: For posterity:  This was discussed briefly on IRC, and Segher approved with some simplifications and a request to implement a fail/retry check. Here's what I have now. No more assembler check 
> On Thu, 4 Nov 2021 at 20:44, Bill Schmidt wrote:
>
>     For posterity:  This was discussed briefly on IRC, and Segher approved with some
>     simplifications and a request to implement a fail/retry check.
>
>
> Here's what I have now. No more assembler check in configure, and it uses the 64-bit __builtin_darn() and truncates it to 32-bit, or retries (up to 100 times) if it fails.
>
> I'm doing some more testing now.
>
Those changes look good from my perspective.  Thanks again for this work!  (Not a maintainer, blah blah...)

Bill
  
Jonathan Wakely Nov. 5, 2021, 6:16 p.m. UTC | #6
On Fri, 5 Nov 2021 at 13:20, Bill Schmidt wrote:

>
> On 11/5/21 7:44 AM, Jonathan Wakely wrote:
> > On Thu, 4 Nov 2021 at 20:44, Bill Schmidt wrote: For posterity:  This
> was discussed briefly on IRC, and Segher approved with some simplifications
> and a request to implement a fail/retry check. Here's what I have now. No
> more assembler check
> > On Thu, 4 Nov 2021 at 20:44, Bill Schmidt wrote:
> >
> >     For posterity:  This was discussed briefly on IRC, and Segher
> approved with some
> >     simplifications and a request to implement a fail/retry check.
> >
> >
> > Here's what I have now. No more assembler check in configure, and it
> uses the 64-bit __builtin_darn() and truncates it to 32-bit, or retries (up
> to 100 times) if it fails.
> >
> > I'm doing some more testing now.
> >
> Those changes look good from my perspective.  Thanks again for this work!
> (Not a maintainer, blah blah...)
>

I changed the number of retries to 10 and pushed it to trunk.
  

Patch

commit a6f925407dd05c593b230da1627435adc53584f8
Author: Jonathan Wakely <jwakely@redhat.com>
Date:   Wed Oct 20 09:25:24 2021

    libstdc++: Add support for POWER9 DARN instruction to std::random_device
    
    The ISA-3.0 instruction set includes DARN ("deliver a random number")
    which can be used similar to the existing support for RDRAND and RDSEED.
    
    libstdc++-v3/ChangeLog:
    
            * acinclude.m4 (GLIBCXX_CHECK_PPC_DARN): Check assembler.
            * config.h.in: Regenerate.
            * configure: Regenerate.
            * configure.ac: Use GLIBCXX_CHECK_PPC_DARN.
            * src/c++11/random.cc [_GLIBCXX_PPC_DARN] (USE_DARN): Define.
            (__ppc_darn): New function to use POWER9 DARN instruction.
            (Which): Add 'darn' enumerator.
            (which_source): Check for __ppc_darn.
            (random_device::_M_init): Support "darn" and "hw" tokens.
            (random_device::_M_getentropy): Add darn to switch.
            * testsuite/26_numerics/random/random_device/cons/token.cc:
            Check "darn" token.
            * testsuite/26_numerics/random/random_device/entropy.cc:
            Likewise.

diff --git a/libstdc++-v3/acinclude.m4 b/libstdc++-v3/acinclude.m4
index 90ecc4a87a2..9ff9ceb20ac 100644
--- a/libstdc++-v3/acinclude.m4
+++ b/libstdc++-v3/acinclude.m4
@@ -4100,6 +4100,27 @@  AC_DEFUN([GLIBCXX_CHECK_X86_RDSEED], [
   AC_MSG_RESULT($ac_cv_x86_rdseed)
 ])
 
+dnl
+dnl Check whether darn is supported in the assembler.
+AC_DEFUN([GLIBCXX_CHECK_PPC_DARN], [
+  AC_MSG_CHECKING([for darn support in assembler])
+  AC_CACHE_VAL(ac_cv_ppc_darn, [
+  ac_cv_ppc_darn=no
+  case "$target" in
+    powerpc*-*-*)
+    AC_TRY_COMPILE(, [
+      signed int x;
+      __asm__ __volatile__ (".machine power9; darn %0,0;": "=r" (x));
+    ], [ac_cv_ppc_darn=yes], [ac_cv_ppc_darn=no])
+  esac
+  ])
+  if test $ac_cv_ppc_darn = yes; then
+    AC_DEFINE(_GLIBCXX_PPC_DARN, 1,
+		[ Defined if as can handle darn. ])
+  fi
+  AC_MSG_RESULT($ac_cv_ppc_darn)
+])
+
 dnl
 dnl Check whether get_nprocs is available in <sys/sysinfo.h>, and define _GLIBCXX_USE_GET_NPROCS.
 dnl
diff --git a/libstdc++-v3/configure.ac b/libstdc++-v3/configure.ac
index 2d68b3672b9..1189c68c380 100644
--- a/libstdc++-v3/configure.ac
+++ b/libstdc++-v3/configure.ac
@@ -467,6 +467,8 @@  GCC_CHECK_ASSEMBLER_HWCAP
 GLIBCXX_CHECK_X86_RDRAND
 # Check if assembler supports rdseed opcode.
 GLIBCXX_CHECK_X86_RDSEED
+# Check if assembler supports darn opcode.
+GLIBCXX_CHECK_PPC_DARN
 
 # This depends on GLIBCXX_ENABLE_SYMVERS and GLIBCXX_IS_NATIVE.
 GLIBCXX_CONFIGURE_TESTSUITE
diff --git a/libstdc++-v3/src/c++11/random.cc b/libstdc++-v3/src/c++11/random.cc
index 4b64bde00ea..213ad691837 100644
--- a/libstdc++-v3/src/c++11/random.cc
+++ b/libstdc++-v3/src/c++11/random.cc
@@ -37,6 +37,9 @@ 
 # ifdef _GLIBCXX_X86_RDSEED
 #  define USE_RDSEED 1
 # endif
+#elif defined __powerpc__ && defined __BUILTIN_CPU_SUPPORTS__ \
+  && defined _GLIBCXX_PPC_DARN
+# define USE_DARN 1
 #endif
 
 #include <cerrno>
@@ -69,7 +72,7 @@ 
 #if defined _GLIBCXX_USE_CRT_RAND_S || defined _GLIBCXX_USE_DEV_RANDOM
 // The OS provides a source of randomness we can use.
 # pragma GCC poison _M_mt
-#elif defined USE_RDRAND || defined USE_RDSEED
+#elif defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
 // Hardware instructions might be available, but use cpuid checks at runtime.
 # pragma GCC poison _M_mt
 // If the runtime cpuid checks fail we'll use a linear congruential engine.
@@ -135,6 +138,15 @@  namespace std _GLIBCXX_VISIBILITY(default)
 #endif
 #endif
 
+#ifdef USE_DARN
+    unsigned int
+    __attribute__((target("cpu=power9")))
+    __ppc_darn(void*)
+    {
+      return __builtin_darn_32();
+    }
+#endif
+
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     unsigned int
     __winxp_rand_s(void*)
@@ -193,11 +205,16 @@  namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif
 
-    enum Which {
-      rand_s = 1, rdseed = 2, rdrand = 4, device_file = 8, prng = 16,
+    enum Which : unsigned {
+      device_file = 1, prng = 2, rand_s = 4,
+      rdseed = 64, rdrand = 128, darn = 256,
       any = 0xffff
     };
 
+    constexpr Which
+    operator|(Which l, Which r) noexcept
+    { return Which(unsigned(l) | unsigned(r)); }
+
     inline Which
     which_source(random_device::result_type (*func [[maybe_unused]])(void*),
 		 void* file [[maybe_unused]])
@@ -221,6 +238,11 @@  namespace std _GLIBCXX_VISIBILITY(default)
 	return rdrand;
 #endif
 
+#ifdef USE_DARN
+      if (func == &__ppc_darn)
+	return darn;
+#endif
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
       if (file != nullptr)
 	return device_file;
@@ -269,6 +291,14 @@  namespace std _GLIBCXX_VISIBILITY(default)
     else if (token == "rdrand" || token == "rdrnd")
       which = rdrand;
 #endif // USE_RDRAND
+#ifdef USE_DARN
+    else if (token == "darn")
+      which = darn;
+#endif
+#if defined USE_RDRAND || defined USE_RDSEED || defined USE_DARN
+    else if (token == "hw" || token == "hardware")
+      which = rdrand | rdseed | darn;
+#endif
 #ifdef _GLIBCXX_USE_CRT_RAND_S
     else if (token == "rand_s")
       which = rand_s;
@@ -346,6 +376,17 @@  namespace std _GLIBCXX_VISIBILITY(default)
     }
 #endif // USE_RDRAND
 
+#ifdef USE_DARN
+    if (which & darn)
+      {
+	if (__builtin_cpu_supports("darn"))
+	  {
+	    _M_func = &__ppc_darn;
+	    return;
+	  }
+      }
+#endif // USE_DARN
+
 #ifdef _GLIBCXX_USE_DEV_RANDOM
     if (which & device_file)
     {
@@ -497,6 +538,7 @@  namespace std _GLIBCXX_VISIBILITY(default)
     {
     case rdrand:
     case rdseed:
+    case darn:
       return (double) max;
     case rand_s:
     case prng:
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
index aeb7403e830..d6ac3a37c64 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/cons/token.cc
@@ -51,8 +51,9 @@  test03()
 {
   // At least one of these tokens should be valid.
   const std::string tokens[] = {
-    "rdseed", "rdrand", "rand_s", "/dev/urandom", "/dev/random", "mt19937",
-    "prng"
+    "rdseed", "rdrand", "darn",
+    "rand_s", "/dev/urandom", "/dev/random",
+    "mt19937", "prng"
   };
   int count = 0;
   for (const std::string& token : tokens)
diff --git a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
index 9ef1538d2bb..6f3ebb1b38e 100644
--- a/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
+++ b/libstdc++-v3/testsuite/26_numerics/random/random_device/entropy.cc
@@ -22,7 +22,7 @@  test01()
       VERIFY( entropy <= max );
     }
 
-  for (auto token : { "rdrand", "rdseed" })
+  for (auto token : { "rdrand", "rdseed", "darn", "hw" })
     if (__gnu_test::random_device_available(token))
     {
       const double entropy = std::random_device(token).entropy();