| Message ID | d1e6867c-7ace-e8c1-5139-e38a249e4398@suse.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | x86emul: remaining AVX512 support | expand |
On 01/07/2019 12:18, Jan Beulich wrote: > There was an encoding mistake in the EVEX Disp8 test code, which was > benign (due to %rdx getting set to zero) to all non-vSIB tests as it > mistakenly encoded <disp8>(%rdx,%rdx) instead of <disp8>(%rdx,%riz). In > the vSIB case this meant <disp8>(%rdx,%zmm2) instead of the intended > <disp8>(%rdx,%zmm4). > > Likewise the access count check wasn't entirely correct for the S/G > case: In the quad-word-index but dword-data case only half the number > of full vector elements get accessed. > > As an unrelated change in the main test harness source file distinguish > the "n/a" messages by bitness. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
--- a/tools/tests/x86_emulator/evex-disp8.c +++ b/tools/tests/x86_emulator/evex-disp8.c @@ -698,7 +698,7 @@ static void test_one(const struct test * instr[3] = evex.raw[2]; instr[4] = test->opc; instr[5] = 0x44 | (test->ext << 3); /* ModR/M */ - instr[6] = 0x12; /* SIB: base rDX, index none / xMM4 */ + instr[6] = 0x22; /* SIB: base rDX, index none / xMM4 */ instr[7] = 1; /* Disp8 */ instr[8] = 0; /* immediate, if any */ @@ -718,7 +718,8 @@ static void test_one(const struct test * if ( accessed[i] ) goto fail; for ( ; i < (test->scale == SC_vl ? vsz : esz) + (sg ? esz : vsz); ++i ) - if ( accessed[i] != (sg ? vsz / esz : 1) ) + if ( accessed[i] != (sg ? (vsz / esz) >> (test->opc & 1 & !evex.w) + : 1) ) goto fail; for ( ; i < ARRAY_SIZE(accessed); ++i ) if ( accessed[i] ) --- a/tools/tests/x86_emulator/test_x86_emulator.c +++ b/tools/tests/x86_emulator/test_x86_emulator.c @@ -4260,7 +4260,7 @@ int main(int argc, char **argv) if ( !blobs[j].size ) { - printf("%-39s n/a\n", blobs[j].name); + printf("%-39s n/a (%u-bit)\n", blobs[j].name, blobs[j].bitness); continue; }
There was an encoding mistake in the EVEX Disp8 test code, which was benign (due to %rdx getting set to zero) to all non-vSIB tests as it mistakenly encoded <disp8>(%rdx,%rdx) instead of <disp8>(%rdx,%riz). In the vSIB case this meant <disp8>(%rdx,%zmm2) instead of the intended <disp8>(%rdx,%zmm4). Likewise the access count check wasn't entirely correct for the S/G case: In the quad-word-index but dword-data case only half the number of full vector elements get accessed. As an unrelated change in the main test harness source file distinguish the "n/a" messages by bitness. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v9: New, split from main gather patch.