From: Mike Larkin <mlarkin@nested.page>
Subject: Re: libm i386: make use of __builtin_cpu_supports()
To: Christian Weisgerber <naddy@mips.inka.de>
Cc: tech@openbsd.org
Date: Sun, 26 Jul 2026 09:22:55 -0700

On Sun, Jul 26, 2026 at 01:05:03AM +0200, Christian Weisgerber wrote:
> On i386, libm queries the machdep.sse sysctl to check at runtime
> whether the CPU supports SSE.
>
> kettenis@ has pointed out elsewhere that the compiler runtime on
> x86 now provides a CPU feature array that we can test with
> __builtin_cpu_supports().  No need to handroll this ourselves.
>
> So let's switch over.  OK?
>

makes sense to me.

> Index: arch/i387/fenv.c
> ===================================================================
> RCS file: /cvs/src/lib/libm/arch/i387/fenv.c,v
> diff -u -p -r1.7 fenv.c
> --- arch/i387/fenv.c	27 Dec 2022 17:10:07 -0000	1.7
> +++ arch/i387/fenv.c	25 Jul 2026 22:17:23 -0000
> @@ -29,7 +29,6 @@
>
>  #include <sys/types.h>
>  #include <sys/time.h>
> -#include <sys/sysctl.h>
>  #include <machine/cpu.h>
>  #include <machine/npx.h>
>
> @@ -63,30 +62,10 @@ fenv_t __fe_dfl_env = {
>  };
>
>  /*
> - * Test for SSE support on this processor.
> - *
>   * We need to use ldmxcsr/stmxcsr to get correct results if any part
>   * of the program was compiled to use SSE floating-point, but we can't
>   * use SSE on older processors.
> - *
> - * In order to do so, we need to query the processor capabilities via the CPUID
> - * instruction. We can make it even simpler though, by querying the machdep.sse
> - * sysctl.
>   */
> -static int __HAS_SSE = 0;
> -
> -static void __test_sse(void) __attribute__ ((constructor));
> -
> -static void __test_sse(void)
> -{
> -	size_t oldlen = sizeof(__HAS_SSE);
> -	int mib[2] = { CTL_MACHDEP, CPU_SSE };
> -	int rv;
> -
> -	rv = sysctl(mib, 2, &__HAS_SSE, &oldlen, NULL, 0);
> -	if (rv == -1)
> -		__HAS_SSE = 0;
> -}
>
>  /*
>   * The feclearexcept() function clears the supported floating-point exceptions
> @@ -110,7 +89,7 @@ feclearexcept(int excepts)
>  	__asm__ volatile ("fldenv %0" : : "m" (fenv));
>
>  	/* Same for SSE environment */
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>  		mxcsr &= ~excepts;
>  		__asm__ volatile ("ldmxcsr %0" : : "m" (mxcsr));
> @@ -137,7 +116,7 @@ fegetexceptflag(fexcept_t *flagp, int ex
>  	__asm__ volatile ("fnstsw %0" : "=am" (status));
>
>  	/* Store the MXCSR register */
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>
>  	/* Store the results in flagp */
> @@ -192,7 +171,7 @@ fesetexceptflag(const fexcept_t *flagp,
>  	__asm__ volatile ("fldenv %0" : : "m" (fenv));
>
>  	/* Same for SSE environment */
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>  		mxcsr &= ~excepts;
>  		mxcsr |= *flagp & excepts;
> @@ -220,7 +199,7 @@ fetestexcept(int excepts)
>  	__asm__ volatile ("fnstsw %0" : "=am" (status));
>
>  	/* Store the MXCSR register state */
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>
>  	return ((status | mxcsr) & excepts);
> @@ -273,7 +252,7 @@ fesetround(int round)
>  	__asm__ volatile ("fldcw %0" : : "m" (control));
>
>  	/* Same for the SSE environment */
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>  		mxcsr &= ~(_X87_ROUND_MASK << _SSE_ROUND_SHIFT);
>  		mxcsr |= round << _SSE_ROUND_SHIFT;
> @@ -295,7 +274,7 @@ fegetenv(fenv_t *envp)
>  	__asm__ volatile ("fnstenv %0" : "=m" (*envp));
>
>  	/* Store the MXCSR register state */
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (envp->__mxcsr));
>
>  	/*
> @@ -329,7 +308,7 @@ feholdexcept(fenv_t *envp)
>  	/* Clear all exception flags in FPU */
>  	__asm__ volatile ("fnclex");
>
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		/* Store the MXCSR register state */
>  		__asm__ volatile ("stmxcsr %0" : "=m" (envp->__mxcsr));
>
> @@ -363,7 +342,7 @@ fesetenv(const fenv_t *envp)
>  	__asm__ volatile ("fldenv %0" : : "m" (*envp));
>
>  	/* Store the MXCSR register */
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("ldmxcsr %0" : : "m" (envp->__mxcsr));
>
>  	return (0);
> @@ -388,7 +367,7 @@ feupdateenv(const fenv_t *envp)
>  	__asm__ volatile ("fnstsw %0" : "=am" (status));
>
>  	/* Store the MXCSR register */
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>
>  	/* Install new floating-point environment */
> @@ -413,14 +392,14 @@ feenableexcept(int mask)
>  	mask &= FE_ALL_EXCEPT;
>
>  	__asm__ volatile ("fnstcw %0" : "=m" (control));
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>
>  	omask = ~(control | (mxcsr >> _SSE_MASK_SHIFT)) & FE_ALL_EXCEPT;
>  	control &= ~mask;
>  	__asm__ volatile ("fldcw %0" : : "m" (control));
>
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		mxcsr &= ~(mask << _SSE_MASK_SHIFT);
>  		__asm__ volatile ("ldmxcsr %0" : : "m" (mxcsr));
>  	}
> @@ -437,14 +416,14 @@ fedisableexcept(int mask)
>  	mask &= FE_ALL_EXCEPT;
>
>  	__asm__ volatile ("fnstcw %0" : "=m" (control));
> -	if (__HAS_SSE)
> +	if (__builtin_cpu_supports("sse"))
>  		__asm__ volatile ("stmxcsr %0" : "=m" (mxcsr));
>
>  	omask = ~(control | (mxcsr >> _SSE_MASK_SHIFT)) & FE_ALL_EXCEPT;
>  	control |= mask;
>  	__asm__ volatile ("fldcw %0" : : "m" (control));
>
> -	if (__HAS_SSE) {
> +	if (__builtin_cpu_supports("sse")) {
>  		mxcsr |= mask << _SSE_MASK_SHIFT;
>  		__asm__ volatile ("ldmxcsr %0" : : "m" (mxcsr));
>  	}
>
> --
> Christian "naddy" Weisgerber                          naddy@mips.inka.de
>