ARM: 8026/1: Fix emulation of multiply accumulate instructions
authorJay Foad <jay.foad@gmail.com>
Mon, 14 Apr 2014 15:23:15 +0000 (16:23 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Mon, 14 Apr 2014 22:28:24 +0000 (23:28 +0100)
The emulation for single and double precision multiply accumulate
instructions correctly normalised any denormal values in the operand
registers, but failed to normalise the destination (accumulator)
register.

This fixes https://bugzilla.kernel.org/show_bug.cgi?id=70501

Signed-off-by: Jay Foad <jay.foad@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
arch/arm/vfp/vfpdouble.c
arch/arm/vfp/vfpsingle.c

index 6cac43b..423f56d 100644 (file)
@@ -866,6 +866,8 @@ vfp_double_multiply_accumulate(int dd, int dn, int dm, u32 fpscr, u32 negate, ch
                vdp.sign = vfp_sign_negate(vdp.sign);
 
        vfp_double_unpack(&vdn, vfp_get_double(dd));
+       if (vdn.exponent == 0 && vdn.significand)
+               vfp_double_normalise_denormal(&vdn);
        if (negate & NEG_SUBTRACT)
                vdn.sign = vfp_sign_negate(vdn.sign);
 
index b252631..4f96c16 100644 (file)
@@ -915,6 +915,8 @@ vfp_single_multiply_accumulate(int sd, int sn, s32 m, u32 fpscr, u32 negate, cha
        v = vfp_get_float(sd);
        pr_debug("VFP: s%u = %08x\n", sd, v);
        vfp_single_unpack(&vsn, v);
+       if (vsn.exponent == 0 && vsn.significand)
+               vfp_single_normalise_denormal(&vsn);
        if (negate & NEG_SUBTRACT)
                vsn.sign = vfp_sign_negate(vsn.sign);