David Howells | 96433f6e | 2012-12-17 13:47:09 +0000 | [diff] [blame] | 1 | #ifndef _UAPI__ASM_ALPHA_FPU_H |
| 2 | #define _UAPI__ASM_ALPHA_FPU_H |
| 3 | |
| 4 | |
| 5 | /* |
| 6 | * Alpha floating-point control register defines: |
| 7 | */ |
| 8 | #define FPCR_DNOD (1UL<<47) /* denorm INV trap disable */ |
| 9 | #define FPCR_DNZ (1UL<<48) /* denorms to zero */ |
| 10 | #define FPCR_INVD (1UL<<49) /* invalid op disable (opt.) */ |
| 11 | #define FPCR_DZED (1UL<<50) /* division by zero disable (opt.) */ |
| 12 | #define FPCR_OVFD (1UL<<51) /* overflow disable (optional) */ |
| 13 | #define FPCR_INV (1UL<<52) /* invalid operation */ |
| 14 | #define FPCR_DZE (1UL<<53) /* division by zero */ |
| 15 | #define FPCR_OVF (1UL<<54) /* overflow */ |
| 16 | #define FPCR_UNF (1UL<<55) /* underflow */ |
| 17 | #define FPCR_INE (1UL<<56) /* inexact */ |
| 18 | #define FPCR_IOV (1UL<<57) /* integer overflow */ |
| 19 | #define FPCR_UNDZ (1UL<<60) /* underflow to zero (opt.) */ |
| 20 | #define FPCR_UNFD (1UL<<61) /* underflow disable (opt.) */ |
| 21 | #define FPCR_INED (1UL<<62) /* inexact disable (opt.) */ |
| 22 | #define FPCR_SUM (1UL<<63) /* summary bit */ |
| 23 | |
| 24 | #define FPCR_DYN_SHIFT 58 /* first dynamic rounding mode bit */ |
| 25 | #define FPCR_DYN_CHOPPED (0x0UL << FPCR_DYN_SHIFT) /* towards 0 */ |
| 26 | #define FPCR_DYN_MINUS (0x1UL << FPCR_DYN_SHIFT) /* towards -INF */ |
| 27 | #define FPCR_DYN_NORMAL (0x2UL << FPCR_DYN_SHIFT) /* towards nearest */ |
| 28 | #define FPCR_DYN_PLUS (0x3UL << FPCR_DYN_SHIFT) /* towards +INF */ |
| 29 | #define FPCR_DYN_MASK (0x3UL << FPCR_DYN_SHIFT) |
| 30 | |
| 31 | #define FPCR_MASK 0xffff800000000000L |
| 32 | |
| 33 | /* |
| 34 | * IEEE trap enables are implemented in software. These per-thread |
| 35 | * bits are stored in the "ieee_state" field of "struct thread_info". |
| 36 | * Thus, the bits are defined so as not to conflict with the |
| 37 | * floating-point enable bit (which is architected). On top of that, |
| 38 | * we want to make these bits compatible with OSF/1 so |
| 39 | * ieee_set_fp_control() etc. can be implemented easily and |
| 40 | * compatibly. The corresponding definitions are in |
| 41 | * /usr/include/machine/fpu.h under OSF/1. |
| 42 | */ |
| 43 | #define IEEE_TRAP_ENABLE_INV (1UL<<1) /* invalid op */ |
| 44 | #define IEEE_TRAP_ENABLE_DZE (1UL<<2) /* division by zero */ |
| 45 | #define IEEE_TRAP_ENABLE_OVF (1UL<<3) /* overflow */ |
| 46 | #define IEEE_TRAP_ENABLE_UNF (1UL<<4) /* underflow */ |
| 47 | #define IEEE_TRAP_ENABLE_INE (1UL<<5) /* inexact */ |
| 48 | #define IEEE_TRAP_ENABLE_DNO (1UL<<6) /* denorm */ |
| 49 | #define IEEE_TRAP_ENABLE_MASK (IEEE_TRAP_ENABLE_INV | IEEE_TRAP_ENABLE_DZE |\ |
| 50 | IEEE_TRAP_ENABLE_OVF | IEEE_TRAP_ENABLE_UNF |\ |
| 51 | IEEE_TRAP_ENABLE_INE | IEEE_TRAP_ENABLE_DNO) |
| 52 | |
| 53 | /* Denorm and Underflow flushing */ |
| 54 | #define IEEE_MAP_DMZ (1UL<<12) /* Map denorm inputs to zero */ |
| 55 | #define IEEE_MAP_UMZ (1UL<<13) /* Map underflowed outputs to zero */ |
| 56 | |
| 57 | #define IEEE_MAP_MASK (IEEE_MAP_DMZ | IEEE_MAP_UMZ) |
| 58 | |
| 59 | /* status bits coming from fpcr: */ |
| 60 | #define IEEE_STATUS_INV (1UL<<17) |
| 61 | #define IEEE_STATUS_DZE (1UL<<18) |
| 62 | #define IEEE_STATUS_OVF (1UL<<19) |
| 63 | #define IEEE_STATUS_UNF (1UL<<20) |
| 64 | #define IEEE_STATUS_INE (1UL<<21) |
| 65 | #define IEEE_STATUS_DNO (1UL<<22) |
| 66 | |
| 67 | #define IEEE_STATUS_MASK (IEEE_STATUS_INV | IEEE_STATUS_DZE | \ |
| 68 | IEEE_STATUS_OVF | IEEE_STATUS_UNF | \ |
| 69 | IEEE_STATUS_INE | IEEE_STATUS_DNO) |
| 70 | |
| 71 | #define IEEE_SW_MASK (IEEE_TRAP_ENABLE_MASK | \ |
| 72 | IEEE_STATUS_MASK | IEEE_MAP_MASK) |
| 73 | |
| 74 | #define IEEE_CURRENT_RM_SHIFT 32 |
| 75 | #define IEEE_CURRENT_RM_MASK (3UL<<IEEE_CURRENT_RM_SHIFT) |
| 76 | |
| 77 | #define IEEE_STATUS_TO_EXCSUM_SHIFT 16 |
| 78 | |
| 79 | #define IEEE_INHERIT (1UL<<63) /* inherit on thread create? */ |
| 80 | |
| 81 | /* |
| 82 | * Convert the software IEEE trap enable and status bits into the |
| 83 | * hardware fpcr format. |
| 84 | * |
| 85 | * Digital Unix engineers receive my thanks for not defining the |
| 86 | * software bits identical to the hardware bits. The chip designers |
| 87 | * receive my thanks for making all the not-implemented fpcr bits |
| 88 | * RAZ forcing us to use system calls to read/write this value. |
| 89 | */ |
| 90 | |
| 91 | static inline unsigned long |
| 92 | ieee_swcr_to_fpcr(unsigned long sw) |
| 93 | { |
| 94 | unsigned long fp; |
| 95 | fp = (sw & IEEE_STATUS_MASK) << 35; |
| 96 | fp |= (sw & IEEE_MAP_DMZ) << 36; |
| 97 | fp |= (sw & IEEE_STATUS_MASK ? FPCR_SUM : 0); |
| 98 | fp |= (~sw & (IEEE_TRAP_ENABLE_INV |
| 99 | | IEEE_TRAP_ENABLE_DZE |
| 100 | | IEEE_TRAP_ENABLE_OVF)) << 48; |
| 101 | fp |= (~sw & (IEEE_TRAP_ENABLE_UNF | IEEE_TRAP_ENABLE_INE)) << 57; |
| 102 | fp |= (sw & IEEE_MAP_UMZ ? FPCR_UNDZ | FPCR_UNFD : 0); |
| 103 | fp |= (~sw & IEEE_TRAP_ENABLE_DNO) << 41; |
| 104 | return fp; |
| 105 | } |
| 106 | |
| 107 | static inline unsigned long |
| 108 | ieee_fpcr_to_swcr(unsigned long fp) |
| 109 | { |
| 110 | unsigned long sw; |
| 111 | sw = (fp >> 35) & IEEE_STATUS_MASK; |
| 112 | sw |= (fp >> 36) & IEEE_MAP_DMZ; |
| 113 | sw |= (~fp >> 48) & (IEEE_TRAP_ENABLE_INV |
| 114 | | IEEE_TRAP_ENABLE_DZE |
| 115 | | IEEE_TRAP_ENABLE_OVF); |
| 116 | sw |= (~fp >> 57) & (IEEE_TRAP_ENABLE_UNF | IEEE_TRAP_ENABLE_INE); |
| 117 | sw |= (fp >> 47) & IEEE_MAP_UMZ; |
| 118 | sw |= (~fp >> 41) & IEEE_TRAP_ENABLE_DNO; |
| 119 | return sw; |
| 120 | } |
| 121 | |
| 122 | |
| 123 | #endif /* _UAPI__ASM_ALPHA_FPU_H */ |