shithub: riscv

Info  •  Files  •  Log  •  Branches

amd64: FP: always use enough to fit AVX state and align to 64 bytes

--- a/sys/src/9/pc64/dat.h

+++ b/sys/src/9/pc64/dat.h

@@ -242,7 +242,6 @@

 	u64int	dr7;			/* shadow copy of dr7 */

 	u64int	xcr0;

 	u32int	fpsavesz;

-	u32int	fpalign;

 	void*	vmx;

--- a/sys/src/9/pc64/fpu.c

+++ b/sys/src/9/pc64/fpu.c

@@ -16,8 +16,7 @@

 	uintptr cr4;

 	ulong regs[4];

-	m->fpsavesz = sizeof(FPssestate);

-	m->fpalign = 16;

+	m->fpsavesz = sizeof(FPsave); /* always enough to fit sse+avx */

 	if((m->cpuiddx & (Sse|Fxsr)) == (Sse|Fxsr)){ /* have sse fp? */

 		cr4 = getcr4() | CR4Osfxsr|CR4Oxmmex;

 		putcr4(cr4);

@@ -34,7 +33,6 @@

 			cpuid(0xd, 0, regs);

 			m->fpsavesz = regs[1];

-			m->fpalign = 64;

 			cpuid(0xd, 1, regs);

 			if(regs[0] & Xsaveopt)

--- a/sys/src/9/pc64/main.c

+++ b/sys/src/9/pc64/main.c

@@ -488,7 +488,7 @@

 			up->fpstate |= FPkernel;

 		}

 		while(up->fpslot[index] == nil)

-			up->fpslot[index] = mallocalign(m->fpsavesz, m->fpalign, 0, 0);

+			up->fpslot[index] = mallocalign(m->fpsavesz, FPalign, 0, 0);

 		up->fpsave = up->fpslot[index];

 		up->fpstate = FPactive | (up->fpstate & (FPnouser|FPkernel|FPindexm));

 		break;

@@ -574,7 +574,7 @@

 	case FPinactive	| FPpush:

 	case FPinactive:

 		while(p->fpslot[0] == nil)

-			p->fpslot[0] = mallocalign(m->fpsavesz, m->fpalign, 0, 0);

+			p->fpslot[0] = mallocalign(m->fpsavesz, FPalign, 0, 0);

 		memmove(p->fpsave = p->fpslot[0], up->fpslot[0], m->fpsavesz);

 		p->fpstate = FPinactive;

 	}

--- a/sys/src/9/pc64/mem.h

+++ b/sys/src/9/pc64/mem.h

@@ -26,6 +26,7 @@

 #define	ROUND(s, sz)	(((s)+((sz)-1))&~((sz)-1))

 #define	PGROUND(s)	ROUND(s, BY2PG)

 #define	BLOCKALIGN	8

+#define	FPalign		64

 #define	MAXMACH		128			/* max # cpus system can run */