This undoes a change to md_assemble() that
32930e4edbc0 ("x86: Support
GNU_PROPERTY_X86_ISA_1_V[234] marker") did without any explanation. This
broke a CVTPI2PS property test that a subsequent test will add, and the
updates to existing tests also demonstrate what was wrong: For example,
AVX insns update the full YMM, even if a Vex128 variant is in use.
+2021-02-16 Jan Beulich <jbeulich@suse.com>
+
+ * config/tc-i386.c (md_assemble): Use template rather than
+ actuals when updating i.xstate.
+ * testsuite/gas/i386/property-4.d,
+ testsuite/gas/i386/property-5.d,
+ testsuite/gas/i386/property-12.d: Adjust expectations.
+
2021-02-16 Jan Beulich <jbeulich@suse.com>
* config/tc-i386.c (output_insn): Handle ldmxcsr, stmxcsr,
for (j = 0; j < i.operands; j++)
{
i.types[j] = operand_type_and (i.types[j], i.tm.operand_types[j]);
- switch (i.types[j].bitfield.class)
+ switch (i.tm.operand_types[j].bitfield.class)
{
default:
break;
i.xstate |= xstate_mask;
break;
case RegSIMD:
- if (i.types[j].bitfield.tmmword)
+ if (i.tm.operand_types[j].bitfield.tmmword)
i.xstate |= xstate_tmm;
- else if (i.types[j].bitfield.zmmword)
+ else if (i.tm.operand_types[j].bitfield.zmmword)
i.xstate |= xstate_zmm;
- else if (i.types[j].bitfield.ymmword)
+ else if (i.tm.operand_types[j].bitfield.ymmword)
i.xstate |= xstate_ymm;
- else if (i.types[j].bitfield.xmmword)
+ else if (i.tm.operand_types[j].bitfield.xmmword)
i.xstate |= xstate_xmm;
break;
}
[ ]+Owner[ ]+Data size[ ]+Description
GNU 0x[0-9a-f]+ NT_GNU_PROPERTY_TYPE_0
Properties: x86 ISA used: x86-64-baseline, x86-64-v4
- x86 feature used: x86, XMM, MASK
+ x86 feature used: x86, XMM, YMM, ZMM, MASK
[ ]+Owner[ ]+Data size[ ]+Description
GNU 0x[0-9a-f]+ NT_GNU_PROPERTY_TYPE_0
Properties: x86 ISA used: x86-64-baseline, x86-64-v3
- x86 feature used: x86, XMM
+ x86 feature used: x86, XMM, YMM
[ ]+Owner[ ]+Data size[ ]+Description
GNU 0x[0-9a-f]+ NT_GNU_PROPERTY_TYPE_0
Properties: x86 ISA used: x86-64-baseline, x86-64-v4
- x86 feature used: x86, XMM
+ x86 feature used: x86, XMM, YMM, ZMM