Re: Benchmark performance

[Atsushi Nemoto <nemoto@toshiba-tops.co.jp>]

>>>>> On Mon, 13 Aug 2001 10:34:46 -0700 (PDT), Wayne Gowcher <wgowcher@yahoo.com> said:
wgowcher> a 23 % reduction in the Floating Point Index benchmark

Current CVS kernel uses FPU emulator unconditionally.  If one floating
point intruction causes a 'Unimplemented' exception (denormalized
result, etc.) following floating point instructions are also handle by
FPU emulator (not only the instruction which raise the exception).

I do not know this is really desired behavior, but here is a patch to
change this.  If Unimplemented exception had been occured during the
benchmark, aplying this patch may result better performance.

---
Atsushi Nemoto
diff -ur linux.sgi/arch/mips/kernel/traps.c linux/arch/mips/kernel/traps.c
--- linux.sgi/arch/mips/kernel/traps.c	Sun Aug  5 23:39:20 2001
+++ linux/arch/mips/kernel/traps.c	Thu Aug 16 12:20:23 2001
@@ -60,7 +60,7 @@
 extern asmlinkage void handle_mcheck(void);
 extern asmlinkage void handle_reserved(void);
 
-extern int fpu_emulator_cop1Handler(struct pt_regs *);
+extern int fpu_emulator_cop1Handler(struct pt_regs *, int);
 
 char watch_available = 0;
 
@@ -306,7 +306,8 @@
 		save_fp(current);
 	
 		/* Run the emulator */
-		sig = fpu_emulator_cop1Handler(regs);
+		/* Emulate only one insn if we have FPU. */
+		sig = fpu_emulator_cop1Handler(regs, 1);
 
 		/* 
 		 * We can't allow the emulated instruction to leave the
@@ -605,7 +606,7 @@
 			current->used_math = 1;
 		}
 	}
-	sig = fpu_emulator_cop1Handler(regs);
+	sig = fpu_emulator_cop1Handler(regs, 0);
 	last_task_used_math = current;
 	if (sig)
 		force_sig(sig, current);
diff -ur linux.sgi/arch/mips/math-emu/cp1emu.c linux/arch/mips/math-emu/cp1emu.c
--- linux.sgi/arch/mips/math-emu/cp1emu.c	Sun Aug  5 23:39:27 2001
+++ linux/arch/mips/math-emu/cp1emu.c	Thu Aug 16 12:21:07 2001
@@ -1662,7 +1662,7 @@
  * hit a non-fp instruction, or a backward branch.  This cuts down dramatically
  * on the per instruction exception overhead.
  */
-int fpu_emulator_cop1Handler(struct pt_regs *xcp)
+int fpu_emulator_cop1Handler(struct pt_regs *xcp, int maxcount)
 {
 	struct mips_fpu_soft_struct *ctx = &current->thread.fpu.soft;
 	unsigned long oldepc, prevepc;
@@ -1682,6 +1682,8 @@
 			sig = cop1Emulate(xcp, ctx);
 		else
 			xcp->cp0_epc += 4;	/* skip nops */
+		if (maxcount && --maxcount <= 0)
+			break;
 	} while (xcp->cp0_epc > prevepc && sig == 0);
 
 	/* SIGILL indicates a non-fpu instruction */