VirtualBox

Changeset 104809 in vbox for trunk/src/VBox/ValidationKit


Ignore:
Timestamp:
May 29, 2024 8:22:23 AM (6 months ago)
Author:
vboxsync
Message:

ValidationKit/bootsectors: bugref:10658 Figured out why SSE instructions worked while AVX did not for certain cases.

File:
1 edited

Legend:

Unmodified
Added
Removed

  • trunk/src/VBox/ValidationKit/bootsectors/bs3-cpu-instr-4.c32

    r104785 r104809  
    514514    uint64_t   fFlags;
    515515    uint16_t   cb       = Bs3ExtCtxGetSize(&fFlags);
    516     PBS3EXTCTX pExtCtx1 = Bs3MemAlloc(BS3MEMKIND_TILED, cb * 3);
     516    PBS3EXTCTX pExtCtx1 = Bs3MemAlloc(BS3MEMKIND_TILED, cb * 2);
    517517    PBS3EXTCTX pExtCtx2 = (PBS3EXTCTX)((uint8_t BS3_FAR *)pExtCtx1 + cb);
    518518    if (pExtCtx1)
    … …  
    10111011    /*
    10121012     * Check the result.
     1013     *
     1014     * If a floating-point exception is expected, the destination is not updated by the instruction.
     1015     * In the case of SSE instructions, updating the destination here will work because it is the same
     1016     * as the source, but for AVX++ it won't because the destination is different and would contain 0s.
    10131017     */
    10141018    cErrors = Bs3TestSubErrorCount();
    10151019    if (   bXcptExpect == X86_XCPT_DB
     1020        && !fFpFlagsExpect
    10161021        && pTest->iRegDst != UINT8_MAX)
    10171022    {
    … …  
    13561361     * Infinity.
    13571362     */
    1358 #if 0
    13591363    /* 5*/{ { /*src2     */ { BS3_FP64_INF(0), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    13601364            { /*src1     */ { BS3_FP64_INF(1), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    … …  
    13751379              /*daz,fz,rc*/ 1, 1, X86_MXCSR_RC_UP,
    13761380              /*flags    */ X86_MXCSR_IE },
    1377 #endif
    13781381    /* 8*/{ { /*src2     */ { BS3_FP64_INF(0),  BS3_FP64_INF(1),  BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    13791382            { /*src1     */ { BS3_FP64_INF(1),  BS3_FP64_INF(0),  BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    … …  
    13851388     * Overflow.
    13861389     */
    1387 #if 0
    13881390    /* 9*/{ { /*src2     */ { BS3_FP64_ZERO(0), BS3_FP64_NORMAL_MAX(1), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    13891391            { /*src1     */ { BS3_FP64_ZERO(0), BS3_FP64_NORMAL_MAX(1), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    … …  
    13981400              /*daz,fz,rc*/ 0, 0, X86_MXCSR_RC_NEAREST,
    13991401              /*flags    */ X86_MXCSR_OE },
    1400 #endif
    14011402    /*11*/{ { /*src2     */ { BS3_FP64_NORMAL_MAX(0), BS3_FP64_NORMAL_MIN(1), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    14021403            { /*src1     */ { BS3_FP64_NORMAL_MAX(0), BS3_FP64_NORMAL_MIN(1), BS3_FP64_ZERO(0), BS3_FP64_ZERO(0) } },
    … …  
    14581459
    14591460        { bs3CpuInstrX_vaddpd_XMM1_XMM2_XMM3_icebp_c16,  X86_XCPT_GP, RM_REG, T_AVX_128, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1460         { bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c16, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1461 
    1462         { bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c16,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1463         { bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c16, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1461        //{ bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c16, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1462        //
     1463        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c16,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1464        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c16, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14641465    };
    14651466    static BS3CPUINSTR4_TEST1_T const s_aTests32[] =
    … …  
    14681469        { bs3CpuInstrX_addpd_XMM1_FSxBX_icebp_c32, 255, RM_MEM, T_SSE2, 1, 1, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14691470
    1470         { bs3CpuInstrX_vaddpd_XMM1_XMM2_XMM3_icebp_c32,  X86_XCPT_GP, RM_REG, T_AVX_128, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1471         { bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c32, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1472 
    1473         { bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c32,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1474         { bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c32, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1471        //{ bs3CpuInstrX_vaddpd_XMM1_XMM2_XMM3_icebp_c32,  X86_XCPT_GP, RM_REG, T_AVX_128, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1472        //{ bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c32, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1473        //
     1474        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c32,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1475        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c32, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14751476    };
    14761477    static BS3CPUINSTR4_TEST1_T const s_aTests64[] =
    … …  
    14791480        { bs3CpuInstrX_addpd_XMM1_FSxBX_icebp_c64, 255, RM_MEM, T_SSE2, 1, 1, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14801481
    1481         { bs3CpuInstrX_vaddpd_XMM1_XMM2_XMM3_icebp_c64,  X86_XCPT_GP, RM_REG, T_AVX_128, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1482         { bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c64, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1483 
    1484         { bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c64,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1485         { bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c64, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1486 
    1487         { bs3CpuInstrX_addpd_XMM8_XMM9_icebp_c64,  255, RM_REG, T_SSE2, 8, 8, 9,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1488         { bs3CpuInstrX_addpd_XMM8_FSxBX_icebp_c64, 255, RM_MEM, T_SSE2, 8, 8, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    1489 
    1490         { bs3CpuInstrX_vaddpd_YMM8_YMM9_YMM10_icebp_c64, X86_XCPT_GP, RM_REG, T_AVX_256, 8, 9, 10,  RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1482        //{ bs3CpuInstrX_vaddpd_XMM1_XMM2_XMM3_icebp_c64,  X86_XCPT_GP, RM_REG, T_AVX_128, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1483        //{ bs3CpuInstrX_vaddpd_XMM1_XMM2_FSxBX_icebp_c64, X86_XCPT_GP, RM_MEM, T_AVX_128, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1484        //
     1485        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_YMM3_icebp_c64,  X86_XCPT_GP, RM_REG, T_AVX2_256, 1, 2, 3,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1486        //{ bs3CpuInstrX_vaddpd_YMM1_YMM2_FSxBX_icebp_c64, X86_XCPT_GP, RM_MEM, T_AVX2_256, 1, 2, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1487        //
     1488        //{ bs3CpuInstrX_addpd_XMM8_XMM9_icebp_c64,  255, RM_REG, T_SSE2, 8, 8, 9,   RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1489        //{ bs3CpuInstrX_addpd_XMM8_FSxBX_icebp_c64, 255, RM_MEM, T_SSE2, 8, 8, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
     1490        //
     1491        //{ bs3CpuInstrX_vaddpd_YMM8_YMM9_YMM10_icebp_c64, X86_XCPT_GP, RM_REG, T_AVX_256, 8, 9, 10,  RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14911492//        { bs3CpuInstrX_vaddpd_YMM8_YMM9_FSxBX_icebp_c64, X86_XCPT_GP, RM_MEM, T_AVX_256, 8, 9, 255, RT_ELEMENTS(s_aValues), (BS3CPUINSTR4_TEST1_VALUES_T *)s_aValues },
    14921493
Note: See TracChangeset for help on using the changeset viewer.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette