Commit 477023a6038e8f457c604b138198f76d093b9a87
1 parent
bfa1e5cf
Improve single-precision floats load & stores:
as the PowerPC registers only store double-precision floats, use float64_to_float32 & float32_to_float64 to do the appropriate conversion. Implement stfiwx. git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@3280 c046a42c-6fe2-441c-8c8c-71466251a162
Showing
2 changed files
with
87 additions
and
52 deletions
target-ppc/op_mem.h
... | ... | @@ -403,11 +403,30 @@ void OPPROTO glue(glue(glue(op_st, name), _64), MEMSUFFIX) (void) \ |
403 | 403 | } |
404 | 404 | #endif |
405 | 405 | |
406 | +static inline void glue(stfs, MEMSUFFIX) (target_ulong EA, double d) | |
407 | +{ | |
408 | + glue(stfl, MEMSUFFIX)(EA, float64_to_float32(d, &env->fp_status)); | |
409 | +} | |
410 | + | |
411 | +static inline void glue(stfiwx, MEMSUFFIX) (target_ulong EA, double d) | |
412 | +{ | |
413 | + union { | |
414 | + double d; | |
415 | + uint64_t u; | |
416 | + } u; | |
417 | + | |
418 | + /* Store the low order 32 bits without any conversion */ | |
419 | + u.d = d; | |
420 | + glue(stl, MEMSUFFIX)(EA, u.u); | |
421 | +} | |
422 | + | |
406 | 423 | PPC_STF_OP(fd, stfq); |
407 | -PPC_STF_OP(fs, stfl); | |
424 | +PPC_STF_OP(fs, stfs); | |
425 | +PPC_STF_OP(fiwx, stfiwx); | |
408 | 426 | #if defined(TARGET_PPC64) |
409 | 427 | PPC_STF_OP_64(fd, stfq); |
410 | -PPC_STF_OP_64(fs, stfl); | |
428 | +PPC_STF_OP_64(fs, stfs); | |
429 | +PPC_STF_OP_64(fiwx, stfiwx); | |
411 | 430 | #endif |
412 | 431 | |
413 | 432 | static inline void glue(stfqr, MEMSUFFIX) (target_ulong EA, double d) |
... | ... | @@ -429,14 +448,14 @@ static inline void glue(stfqr, MEMSUFFIX) (target_ulong EA, double d) |
429 | 448 | glue(stfq, MEMSUFFIX)(EA, u.d); |
430 | 449 | } |
431 | 450 | |
432 | -static inline void glue(stflr, MEMSUFFIX) (target_ulong EA, float f) | |
451 | +static inline void glue(stfsr, MEMSUFFIX) (target_ulong EA, double d) | |
433 | 452 | { |
434 | 453 | union { |
435 | 454 | float f; |
436 | 455 | uint32_t u; |
437 | 456 | } u; |
438 | 457 | |
439 | - u.f = f; | |
458 | + u.f = float64_to_float32(d, &env->fp_status); | |
440 | 459 | u.u = ((u.u & 0xFF000000UL) >> 24) | |
441 | 460 | ((u.u & 0x00FF0000ULL) >> 8) | |
442 | 461 | ((u.u & 0x0000FF00UL) << 8) | |
... | ... | @@ -444,11 +463,30 @@ static inline void glue(stflr, MEMSUFFIX) (target_ulong EA, float f) |
444 | 463 | glue(stfl, MEMSUFFIX)(EA, u.f); |
445 | 464 | } |
446 | 465 | |
466 | +static inline void glue(stfiwxr, MEMSUFFIX) (target_ulong EA, double d) | |
467 | +{ | |
468 | + union { | |
469 | + double d; | |
470 | + uint64_t u; | |
471 | + } u; | |
472 | + | |
473 | + /* Store the low order 32 bits without any conversion */ | |
474 | + u.d = d; | |
475 | + u.u = ((u.u & 0xFF000000UL) >> 24) | | |
476 | + ((u.u & 0x00FF0000ULL) >> 8) | | |
477 | + ((u.u & 0x0000FF00UL) << 8) | | |
478 | + ((u.u & 0x000000FFULL) << 24); | |
479 | + glue(stl, MEMSUFFIX)(EA, u.u); | |
480 | +} | |
481 | + | |
482 | + | |
447 | 483 | PPC_STF_OP(fd_le, stfqr); |
448 | -PPC_STF_OP(fs_le, stflr); | |
484 | +PPC_STF_OP(fs_le, stfsr); | |
485 | +PPC_STF_OP(fiwx_le, stfiwxr); | |
449 | 486 | #if defined(TARGET_PPC64) |
450 | 487 | PPC_STF_OP_64(fd_le, stfqr); |
451 | -PPC_STF_OP_64(fs_le, stflr); | |
488 | +PPC_STF_OP_64(fs_le, stfsr); | |
489 | +PPC_STF_OP_64(fiwx_le, stfiwxr); | |
452 | 490 | #endif |
453 | 491 | |
454 | 492 | /*** Floating-point load ***/ |
... | ... | @@ -468,11 +506,16 @@ void OPPROTO glue(glue(glue(op_l, name), _64), MEMSUFFIX) (void) \ |
468 | 506 | } |
469 | 507 | #endif |
470 | 508 | |
509 | +static inline double glue(ldfs, MEMSUFFIX) (target_ulong EA) | |
510 | +{ | |
511 | + return float32_to_float64(glue(ldfl, MEMSUFFIX)(EA), &env->fp_status); | |
512 | +} | |
513 | + | |
471 | 514 | PPC_LDF_OP(fd, ldfq); |
472 | -PPC_LDF_OP(fs, ldfl); | |
515 | +PPC_LDF_OP(fs, ldfs); | |
473 | 516 | #if defined(TARGET_PPC64) |
474 | 517 | PPC_LDF_OP_64(fd, ldfq); |
475 | -PPC_LDF_OP_64(fs, ldfl); | |
518 | +PPC_LDF_OP_64(fs, ldfs); | |
476 | 519 | #endif |
477 | 520 | |
478 | 521 | static inline double glue(ldfqr, MEMSUFFIX) (target_ulong EA) |
... | ... | @@ -495,7 +538,7 @@ static inline double glue(ldfqr, MEMSUFFIX) (target_ulong EA) |
495 | 538 | return u.d; |
496 | 539 | } |
497 | 540 | |
498 | -static inline float glue(ldflr, MEMSUFFIX) (target_ulong EA) | |
541 | +static inline double glue(ldfsr, MEMSUFFIX) (target_ulong EA) | |
499 | 542 | { |
500 | 543 | union { |
501 | 544 | float f; |
... | ... | @@ -508,14 +551,14 @@ static inline float glue(ldflr, MEMSUFFIX) (target_ulong EA) |
508 | 551 | ((u.u & 0x0000FF00UL) << 8) | |
509 | 552 | ((u.u & 0x000000FFULL) << 24); |
510 | 553 | |
511 | - return u.f; | |
554 | + return float32_to_float64(u.f, &env->fp_status); | |
512 | 555 | } |
513 | 556 | |
514 | 557 | PPC_LDF_OP(fd_le, ldfqr); |
515 | -PPC_LDF_OP(fs_le, ldflr); | |
558 | +PPC_LDF_OP(fs_le, ldfsr); | |
516 | 559 | #if defined(TARGET_PPC64) |
517 | 560 | PPC_LDF_OP_64(fd_le, ldfqr); |
518 | -PPC_LDF_OP_64(fs_le, ldflr); | |
561 | +PPC_LDF_OP_64(fs_le, ldfsr); | |
519 | 562 | #endif |
520 | 563 | |
521 | 564 | /* Load and set reservation */ | ... | ... |
target-ppc/translate.c
... | ... | @@ -2581,8 +2581,8 @@ GEN_HANDLER(sync, 0x1F, 0x16, 0x12, 0x03CF0801, PPC_MEM_SYNC) |
2581 | 2581 | } |
2582 | 2582 | |
2583 | 2583 | /*** Floating-point load ***/ |
2584 | -#define GEN_LDF(width, opc) \ | |
2585 | -GEN_HANDLER(l##width, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ | |
2584 | +#define GEN_LDF(width, opc, type) \ | |
2585 | +GEN_HANDLER(l##width, opc, 0xFF, 0xFF, 0x00000000, type) \ | |
2586 | 2586 | { \ |
2587 | 2587 | if (unlikely(!ctx->fpu_enabled)) { \ |
2588 | 2588 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2593,8 +2593,8 @@ GEN_HANDLER(l##width, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ |
2593 | 2593 | gen_op_store_FT0_fpr(rD(ctx->opcode)); \ |
2594 | 2594 | } |
2595 | 2595 | |
2596 | -#define GEN_LDUF(width, opc) \ | |
2597 | -GEN_HANDLER(l##width##u, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ | |
2596 | +#define GEN_LDUF(width, opc, type) \ | |
2597 | +GEN_HANDLER(l##width##u, opc, 0xFF, 0xFF, 0x00000000, type) \ | |
2598 | 2598 | { \ |
2599 | 2599 | if (unlikely(!ctx->fpu_enabled)) { \ |
2600 | 2600 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2610,8 +2610,8 @@ GEN_HANDLER(l##width##u, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ |
2610 | 2610 | gen_op_store_T0_gpr(rA(ctx->opcode)); \ |
2611 | 2611 | } |
2612 | 2612 | |
2613 | -#define GEN_LDUXF(width, opc) \ | |
2614 | -GEN_HANDLER(l##width##ux, 0x1F, 0x17, opc, 0x00000001, PPC_FLOAT) \ | |
2613 | +#define GEN_LDUXF(width, opc, type) \ | |
2614 | +GEN_HANDLER(l##width##ux, 0x1F, 0x17, opc, 0x00000001, type) \ | |
2615 | 2615 | { \ |
2616 | 2616 | if (unlikely(!ctx->fpu_enabled)) { \ |
2617 | 2617 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2627,8 +2627,8 @@ GEN_HANDLER(l##width##ux, 0x1F, 0x17, opc, 0x00000001, PPC_FLOAT) \ |
2627 | 2627 | gen_op_store_T0_gpr(rA(ctx->opcode)); \ |
2628 | 2628 | } |
2629 | 2629 | |
2630 | -#define GEN_LDXF(width, opc2, opc3) \ | |
2631 | -GEN_HANDLER(l##width##x, 0x1F, opc2, opc3, 0x00000001, PPC_FLOAT) \ | |
2630 | +#define GEN_LDXF(width, opc2, opc3, type) \ | |
2631 | +GEN_HANDLER(l##width##x, 0x1F, opc2, opc3, 0x00000001, type) \ | |
2632 | 2632 | { \ |
2633 | 2633 | if (unlikely(!ctx->fpu_enabled)) { \ |
2634 | 2634 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2639,21 +2639,21 @@ GEN_HANDLER(l##width##x, 0x1F, opc2, opc3, 0x00000001, PPC_FLOAT) \ |
2639 | 2639 | gen_op_store_FT0_fpr(rD(ctx->opcode)); \ |
2640 | 2640 | } |
2641 | 2641 | |
2642 | -#define GEN_LDFS(width, op) \ | |
2642 | +#define GEN_LDFS(width, op, type) \ | |
2643 | 2643 | OP_LD_TABLE(width); \ |
2644 | -GEN_LDF(width, op | 0x20); \ | |
2645 | -GEN_LDUF(width, op | 0x21); \ | |
2646 | -GEN_LDUXF(width, op | 0x01); \ | |
2647 | -GEN_LDXF(width, 0x17, op | 0x00) | |
2644 | +GEN_LDF(width, op | 0x20, type); \ | |
2645 | +GEN_LDUF(width, op | 0x21, type); \ | |
2646 | +GEN_LDUXF(width, op | 0x01, type); \ | |
2647 | +GEN_LDXF(width, 0x17, op | 0x00, type) | |
2648 | 2648 | |
2649 | 2649 | /* lfd lfdu lfdux lfdx */ |
2650 | -GEN_LDFS(fd, 0x12); | |
2650 | +GEN_LDFS(fd, 0x12, PPC_FLOAT); | |
2651 | 2651 | /* lfs lfsu lfsux lfsx */ |
2652 | -GEN_LDFS(fs, 0x10); | |
2652 | +GEN_LDFS(fs, 0x10, PPC_FLOAT); | |
2653 | 2653 | |
2654 | 2654 | /*** Floating-point store ***/ |
2655 | -#define GEN_STF(width, opc) \ | |
2656 | -GEN_HANDLER(st##width, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ | |
2655 | +#define GEN_STF(width, opc, type) \ | |
2656 | +GEN_HANDLER(st##width, opc, 0xFF, 0xFF, 0x00000000, type) \ | |
2657 | 2657 | { \ |
2658 | 2658 | if (unlikely(!ctx->fpu_enabled)) { \ |
2659 | 2659 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2664,8 +2664,8 @@ GEN_HANDLER(st##width, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ |
2664 | 2664 | op_ldst(st##width); \ |
2665 | 2665 | } |
2666 | 2666 | |
2667 | -#define GEN_STUF(width, opc) \ | |
2668 | -GEN_HANDLER(st##width##u, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ | |
2667 | +#define GEN_STUF(width, opc, type) \ | |
2668 | +GEN_HANDLER(st##width##u, opc, 0xFF, 0xFF, 0x00000000, type) \ | |
2669 | 2669 | { \ |
2670 | 2670 | if (unlikely(!ctx->fpu_enabled)) { \ |
2671 | 2671 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2681,8 +2681,8 @@ GEN_HANDLER(st##width##u, opc, 0xFF, 0xFF, 0x00000000, PPC_FLOAT) \ |
2681 | 2681 | gen_op_store_T0_gpr(rA(ctx->opcode)); \ |
2682 | 2682 | } |
2683 | 2683 | |
2684 | -#define GEN_STUXF(width, opc) \ | |
2685 | -GEN_HANDLER(st##width##ux, 0x1F, 0x17, opc, 0x00000001, PPC_FLOAT) \ | |
2684 | +#define GEN_STUXF(width, opc, type) \ | |
2685 | +GEN_HANDLER(st##width##ux, 0x1F, 0x17, opc, 0x00000001, type) \ | |
2686 | 2686 | { \ |
2687 | 2687 | if (unlikely(!ctx->fpu_enabled)) { \ |
2688 | 2688 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2698,8 +2698,8 @@ GEN_HANDLER(st##width##ux, 0x1F, 0x17, opc, 0x00000001, PPC_FLOAT) \ |
2698 | 2698 | gen_op_store_T0_gpr(rA(ctx->opcode)); \ |
2699 | 2699 | } |
2700 | 2700 | |
2701 | -#define GEN_STXF(width, opc2, opc3) \ | |
2702 | -GEN_HANDLER(st##width##x, 0x1F, opc2, opc3, 0x00000001, PPC_FLOAT) \ | |
2701 | +#define GEN_STXF(width, opc2, opc3, type) \ | |
2702 | +GEN_HANDLER(st##width##x, 0x1F, opc2, opc3, 0x00000001, type) \ | |
2703 | 2703 | { \ |
2704 | 2704 | if (unlikely(!ctx->fpu_enabled)) { \ |
2705 | 2705 | GEN_EXCP_NO_FP(ctx); \ |
... | ... | @@ -2710,30 +2710,22 @@ GEN_HANDLER(st##width##x, 0x1F, opc2, opc3, 0x00000001, PPC_FLOAT) \ |
2710 | 2710 | op_ldst(st##width); \ |
2711 | 2711 | } |
2712 | 2712 | |
2713 | -#define GEN_STFS(width, op) \ | |
2713 | +#define GEN_STFS(width, op, type) \ | |
2714 | 2714 | OP_ST_TABLE(width); \ |
2715 | -GEN_STF(width, op | 0x20); \ | |
2716 | -GEN_STUF(width, op | 0x21); \ | |
2717 | -GEN_STUXF(width, op | 0x01); \ | |
2718 | -GEN_STXF(width, 0x17, op | 0x00) | |
2715 | +GEN_STF(width, op | 0x20, type); \ | |
2716 | +GEN_STUF(width, op | 0x21, type); \ | |
2717 | +GEN_STUXF(width, op | 0x01, type); \ | |
2718 | +GEN_STXF(width, 0x17, op | 0x00, type) | |
2719 | 2719 | |
2720 | 2720 | /* stfd stfdu stfdux stfdx */ |
2721 | -GEN_STFS(fd, 0x16); | |
2721 | +GEN_STFS(fd, 0x16, PPC_FLOAT); | |
2722 | 2722 | /* stfs stfsu stfsux stfsx */ |
2723 | -GEN_STFS(fs, 0x14); | |
2723 | +GEN_STFS(fs, 0x14, PPC_FLOAT); | |
2724 | 2724 | |
2725 | 2725 | /* Optional: */ |
2726 | 2726 | /* stfiwx */ |
2727 | -GEN_HANDLER(stfiwx, 0x1F, 0x17, 0x1E, 0x00000001, PPC_FLOAT_STFIWX) | |
2728 | -{ | |
2729 | - if (unlikely(!ctx->fpu_enabled)) { | |
2730 | - GEN_EXCP_NO_FP(ctx); | |
2731 | - return; | |
2732 | - } | |
2733 | - gen_addr_reg_index(ctx); | |
2734 | - /* XXX: TODO: memcpy low order 32 bits of FRP(rs) into memory */ | |
2735 | - GEN_EXCP_INVAL(ctx); | |
2736 | -} | |
2727 | +OP_ST_TABLE(fiwx); | |
2728 | +GEN_STXF(fiwx, 0x17, 0x1E, PPC_FLOAT_STFIWX); | |
2737 | 2729 | |
2738 | 2730 | /*** Branch ***/ |
2739 | 2731 | static inline void gen_goto_tb (DisasContext *ctx, int n, target_ulong dest) | ... | ... |