diff options
| -rw-r--r-- | Documentation/dontdiff | 1 | ||||
| -rw-r--r-- | arch/arm/mach-ixp4xx/coyote-setup.c | 2 | ||||
| -rw-r--r-- | arch/arm/mach-ixp4xx/gtwx5715-setup.c | 2 | ||||
| -rw-r--r-- | arch/arm/mach-ixp4xx/ixdp425-setup.c | 4 | ||||
| -rw-r--r-- | arch/arm/mach-s3c2410/mach-bast.c | 16 | ||||
| -rw-r--r-- | arch/arm/mm/proc-xscale.S | 136 | ||||
| -rw-r--r-- | arch/arm/nwfpe/double_cpdo.c | 24 | ||||
| -rw-r--r-- | arch/arm/nwfpe/extended_cpdo.c | 24 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpa11.c | 30 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpa11.h | 11 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpa11_cpdo.c | 28 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpa11_cpdt.c | 22 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpa11_cprt.c | 28 | ||||
| -rw-r--r-- | arch/arm/nwfpe/fpmodule.c | 15 | ||||
| -rw-r--r-- | arch/arm/nwfpe/single_cpdo.c | 24 | ||||
| -rw-r--r-- | arch/arm/nwfpe/softfloat.c | 334 | ||||
| -rw-r--r-- | arch/arm/nwfpe/softfloat.h | 68 | ||||
| -rw-r--r-- | arch/arm/vfp/vfpdouble.c | 3 | ||||
| -rw-r--r-- | drivers/char/watchdog/sa1100_wdt.c | 49 |
19 files changed, 337 insertions, 484 deletions
diff --git a/Documentation/dontdiff b/Documentation/dontdiff index b974cf595d01..96bea278bbf6 100644 --- a/Documentation/dontdiff +++ b/Documentation/dontdiff | |||
| @@ -104,6 +104,7 @@ logo_*.c | |||
| 104 | logo_*_clut224.c | 104 | logo_*_clut224.c |
| 105 | logo_*_mono.c | 105 | logo_*_mono.c |
| 106 | lxdialog | 106 | lxdialog |
| 107 | mach-types | ||
| 107 | mach-types.h | 108 | mach-types.h |
| 108 | make_times_h | 109 | make_times_h |
| 109 | map | 110 | map |
diff --git a/arch/arm/mach-ixp4xx/coyote-setup.c b/arch/arm/mach-ixp4xx/coyote-setup.c index 4ff4393ef0ea..7f58afb27e71 100644 --- a/arch/arm/mach-ixp4xx/coyote-setup.c +++ b/arch/arm/mach-ixp4xx/coyote-setup.c | |||
| @@ -61,7 +61,7 @@ static struct plat_serial8250_port coyote_uart_data[] = { | |||
| 61 | .mapbase = IXP4XX_UART2_BASE_PHYS, | 61 | .mapbase = IXP4XX_UART2_BASE_PHYS, |
| 62 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, | 62 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, |
| 63 | .irq = IRQ_IXP4XX_UART2, | 63 | .irq = IRQ_IXP4XX_UART2, |
| 64 | .flags = UPF_BOOT_AUTOCONF, | 64 | .flags = UPF_BOOT_AUTOCONF | UPF_SKIP_TEST, |
| 65 | .iotype = UPIO_MEM, | 65 | .iotype = UPIO_MEM, |
| 66 | .regshift = 2, | 66 | .regshift = 2, |
| 67 | .uartclk = IXP4XX_UART_XTAL, | 67 | .uartclk = IXP4XX_UART_XTAL, |
diff --git a/arch/arm/mach-ixp4xx/gtwx5715-setup.c b/arch/arm/mach-ixp4xx/gtwx5715-setup.c index 8ba1cd9406e7..65e356bd10d6 100644 --- a/arch/arm/mach-ixp4xx/gtwx5715-setup.c +++ b/arch/arm/mach-ixp4xx/gtwx5715-setup.c | |||
| @@ -83,7 +83,7 @@ static struct plat_serial8250_port gtwx5715_uart_platform_data[] = { | |||
| 83 | .mapbase = IXP4XX_UART2_BASE_PHYS, | 83 | .mapbase = IXP4XX_UART2_BASE_PHYS, |
| 84 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, | 84 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, |
| 85 | .irq = IRQ_IXP4XX_UART2, | 85 | .irq = IRQ_IXP4XX_UART2, |
| 86 | .flags = UPF_BOOT_AUTOCONF, | 86 | .flags = UPF_BOOT_AUTOCONF | UPF_SKIP_TEST, |
| 87 | .iotype = UPIO_MEM, | 87 | .iotype = UPIO_MEM, |
| 88 | .regshift = 2, | 88 | .regshift = 2, |
| 89 | .uartclk = IXP4XX_UART_XTAL, | 89 | .uartclk = IXP4XX_UART_XTAL, |
diff --git a/arch/arm/mach-ixp4xx/ixdp425-setup.c b/arch/arm/mach-ixp4xx/ixdp425-setup.c index c2ba759e9946..4633470a6a37 100644 --- a/arch/arm/mach-ixp4xx/ixdp425-setup.c +++ b/arch/arm/mach-ixp4xx/ixdp425-setup.c | |||
| @@ -82,7 +82,7 @@ static struct plat_serial8250_port ixdp425_uart_data[] = { | |||
| 82 | .mapbase = IXP4XX_UART1_BASE_PHYS, | 82 | .mapbase = IXP4XX_UART1_BASE_PHYS, |
| 83 | .membase = (char *)IXP4XX_UART1_BASE_VIRT + REG_OFFSET, | 83 | .membase = (char *)IXP4XX_UART1_BASE_VIRT + REG_OFFSET, |
| 84 | .irq = IRQ_IXP4XX_UART1, | 84 | .irq = IRQ_IXP4XX_UART1, |
| 85 | .flags = UPF_BOOT_AUTOCONF, | 85 | .flags = UPF_BOOT_AUTOCONF | UPF_SKIP_TEST, |
| 86 | .iotype = UPIO_MEM, | 86 | .iotype = UPIO_MEM, |
| 87 | .regshift = 2, | 87 | .regshift = 2, |
| 88 | .uartclk = IXP4XX_UART_XTAL, | 88 | .uartclk = IXP4XX_UART_XTAL, |
| @@ -91,7 +91,7 @@ static struct plat_serial8250_port ixdp425_uart_data[] = { | |||
| 91 | .mapbase = IXP4XX_UART2_BASE_PHYS, | 91 | .mapbase = IXP4XX_UART2_BASE_PHYS, |
| 92 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, | 92 | .membase = (char *)IXP4XX_UART2_BASE_VIRT + REG_OFFSET, |
| 93 | .irq = IRQ_IXP4XX_UART1, | 93 | .irq = IRQ_IXP4XX_UART1, |
| 94 | .flags = UPF_BOOT_AUTOCONF, | 94 | .flags = UPF_BOOT_AUTOCONF | UPF_SKIP_TEST, |
| 95 | .iotype = UPIO_MEM, | 95 | .iotype = UPIO_MEM, |
| 96 | .regshift = 2, | 96 | .regshift = 2, |
| 97 | .uartclk = IXP4XX_UART_XTAL, | 97 | .uartclk = IXP4XX_UART_XTAL, |
diff --git a/arch/arm/mach-s3c2410/mach-bast.c b/arch/arm/mach-s3c2410/mach-bast.c index 1e7f343822d0..e9182242da95 100644 --- a/arch/arm/mach-s3c2410/mach-bast.c +++ b/arch/arm/mach-s3c2410/mach-bast.c | |||
| @@ -30,6 +30,7 @@ | |||
| 30 | * 28-Jun-2005 BJD Moved pm functionality out to common code | 30 | * 28-Jun-2005 BJD Moved pm functionality out to common code |
| 31 | * 17-Jul-2005 BJD Changed to platform device for SuperIO 16550s | 31 | * 17-Jul-2005 BJD Changed to platform device for SuperIO 16550s |
| 32 | * 25-Jul-2005 BJD Removed ASIX static mappings | 32 | * 25-Jul-2005 BJD Removed ASIX static mappings |
| 33 | * 27-Jul-2005 BJD Ensure maximum frequency of i2c bus | ||
| 33 | */ | 34 | */ |
| 34 | 35 | ||
| 35 | #include <linux/kernel.h> | 36 | #include <linux/kernel.h> |
| @@ -60,6 +61,7 @@ | |||
| 60 | #include <asm/arch/regs-mem.h> | 61 | #include <asm/arch/regs-mem.h> |
| 61 | #include <asm/arch/regs-lcd.h> | 62 | #include <asm/arch/regs-lcd.h> |
| 62 | #include <asm/arch/nand.h> | 63 | #include <asm/arch/nand.h> |
| 64 | #include <asm/arch/iic.h> | ||
| 63 | 65 | ||
| 64 | #include <linux/mtd/mtd.h> | 66 | #include <linux/mtd/mtd.h> |
| 65 | #include <linux/mtd/nand.h> | 67 | #include <linux/mtd/nand.h> |
| @@ -304,7 +306,7 @@ static void bast_nand_select(struct s3c2410_nand_set *set, int slot) | |||
| 304 | } | 306 | } |
| 305 | 307 | ||
| 306 | static struct s3c2410_platform_nand bast_nand_info = { | 308 | static struct s3c2410_platform_nand bast_nand_info = { |
| 307 | .tacls = 80, | 309 | .tacls = 40, |
| 308 | .twrph0 = 80, | 310 | .twrph0 = 80, |
| 309 | .twrph1 = 80, | 311 | .twrph1 = 80, |
| 310 | .nr_sets = ARRAY_SIZE(bast_nand_sets), | 312 | .nr_sets = ARRAY_SIZE(bast_nand_sets), |
| @@ -385,6 +387,17 @@ static struct platform_device bast_sio = { | |||
| 385 | }, | 387 | }, |
| 386 | }; | 388 | }; |
| 387 | 389 | ||
| 390 | /* we have devices on the bus which cannot work much over the | ||
| 391 | * standard 100KHz i2c bus frequency | ||
| 392 | */ | ||
| 393 | |||
| 394 | static struct s3c2410_platform_i2c bast_i2c_info = { | ||
| 395 | .flags = 0, | ||
| 396 | .slave_addr = 0x10, | ||
| 397 | .bus_freq = 100*1000, | ||
| 398 | .max_freq = 130*1000, | ||
| 399 | }; | ||
| 400 | |||
| 388 | /* Standard BAST devices */ | 401 | /* Standard BAST devices */ |
| 389 | 402 | ||
| 390 | static struct platform_device *bast_devices[] __initdata = { | 403 | static struct platform_device *bast_devices[] __initdata = { |
| @@ -431,6 +444,7 @@ void __init bast_map_io(void) | |||
| 431 | s3c24xx_uclk.parent = &s3c24xx_clkout1; | 444 | s3c24xx_uclk.parent = &s3c24xx_clkout1; |
| 432 | 445 | ||
| 433 | s3c_device_nand.dev.platform_data = &bast_nand_info; | 446 | s3c_device_nand.dev.platform_data = &bast_nand_info; |
| 447 | s3c_device_i2c.dev.platform_data = &bast_i2c_info; | ||
| 434 | 448 | ||
| 435 | s3c24xx_init_io(bast_iodesc, ARRAY_SIZE(bast_iodesc)); | 449 | s3c24xx_init_io(bast_iodesc, ARRAY_SIZE(bast_iodesc)); |
| 436 | s3c24xx_init_clocks(0); | 450 | s3c24xx_init_clocks(0); |
diff --git a/arch/arm/mm/proc-xscale.S b/arch/arm/mm/proc-xscale.S index 2d977b4eeeab..b88de2700146 100644 --- a/arch/arm/mm/proc-xscale.S +++ b/arch/arm/mm/proc-xscale.S | |||
| @@ -370,142 +370,6 @@ ENTRY(cpu_xscale_dcache_clean_area) | |||
| 370 | bhi 1b | 370 | bhi 1b |
| 371 | mov pc, lr | 371 | mov pc, lr |
| 372 | 372 | ||
| 373 | /* ================================ CACHE LOCKING============================ | ||
| 374 | * | ||
| 375 | * The XScale MicroArchitecture implements support for locking entries into | ||
| 376 | * the data and instruction cache. The following functions implement the core | ||
| 377 | * low level instructions needed to accomplish the locking. The developer's | ||
| 378 | * manual states that the code that performs the locking must be in non-cached | ||
| 379 | * memory. To accomplish this, the code in xscale-cache-lock.c copies the | ||
| 380 | * following functions from the cache into a non-cached memory region that | ||
| 381 | * is allocated through consistent_alloc(). | ||
| 382 | * | ||
| 383 | */ | ||
| 384 | .align 5 | ||
| 385 | /* | ||
| 386 | * xscale_icache_lock | ||
| 387 | * | ||
| 388 | * r0: starting address to lock | ||
| 389 | * r1: end address to lock | ||
| 390 | */ | ||
| 391 | ENTRY(xscale_icache_lock) | ||
| 392 | |||
| 393 | iLockLoop: | ||
| 394 | bic r0, r0, #CACHELINESIZE - 1 | ||
| 395 | mcr p15, 0, r0, c9, c1, 0 @ lock into cache | ||
| 396 | cmp r0, r1 @ are we done? | ||
| 397 | add r0, r0, #CACHELINESIZE @ advance to next cache line | ||
| 398 | bls iLockLoop | ||
| 399 | mov pc, lr | ||
| 400 | |||
| 401 | /* | ||
| 402 | * xscale_icache_unlock | ||
| 403 | */ | ||
| 404 | ENTRY(xscale_icache_unlock) | ||
| 405 | mcr p15, 0, r0, c9, c1, 1 @ Unlock icache | ||
| 406 | mov pc, lr | ||
| 407 | |||
| 408 | /* | ||
| 409 | * xscale_dcache_lock | ||
| 410 | * | ||
| 411 | * r0: starting address to lock | ||
| 412 | * r1: end address to lock | ||
| 413 | */ | ||
| 414 | ENTRY(xscale_dcache_lock) | ||
| 415 | mcr p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer | ||
| 416 | mov r2, #1 | ||
| 417 | mcr p15, 0, r2, c9, c2, 0 @ Put dcache in lock mode | ||
| 418 | cpwait ip @ Wait for completion | ||
| 419 | |||
| 420 | mrs r2, cpsr | ||
| 421 | orr r3, r2, #PSR_F_BIT | PSR_I_BIT | ||
| 422 | dLockLoop: | ||
| 423 | msr cpsr_c, r3 | ||
| 424 | mcr p15, 0, r0, c7, c10, 1 @ Write back line if it is dirty | ||
| 425 | mcr p15, 0, r0, c7, c6, 1 @ Flush/invalidate line | ||
| 426 | msr cpsr_c, r2 | ||
| 427 | ldr ip, [r0], #CACHELINESIZE @ Preload 32 bytes into cache from | ||
| 428 | @ location [r0]. Post-increment | ||
| 429 | @ r3 to next cache line | ||
| 430 | cmp r0, r1 @ Are we done? | ||
| 431 | bls dLockLoop | ||
| 432 | |||
| 433 | mcr p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer | ||
| 434 | mov r2, #0 | ||
| 435 | mcr p15, 0, r2, c9, c2, 0 @ Get out of lock mode | ||
| 436 | cpwait_ret lr, ip | ||
| 437 | |||
| 438 | /* | ||
| 439 | * xscale_dcache_unlock | ||
| 440 | */ | ||
| 441 | ENTRY(xscale_dcache_unlock) | ||
| 442 | mcr p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer | ||
| 443 | mcr p15, 0, ip, c9, c2, 1 @ Unlock cache | ||
| 444 | mov pc, lr | ||
| 445 | |||
| 446 | /* | ||
| 447 | * Needed to determine the length of the code that needs to be copied. | ||
| 448 | */ | ||
| 449 | .align 5 | ||
| 450 | ENTRY(xscale_cache_dummy) | ||
| 451 | mov pc, lr | ||
| 452 | |||
| 453 | /* ================================ TLB LOCKING============================== | ||
| 454 | * | ||
| 455 | * The XScale MicroArchitecture implements support for locking entries into | ||
| 456 | * the Instruction and Data TLBs. The following functions provide the | ||
| 457 | * low level support for supporting these under Linux. xscale-lock.c | ||
| 458 | * implements some higher level management code. Most of the following | ||
| 459 | * is taken straight out of the Developer's Manual. | ||
| 460 | */ | ||
| 461 | |||
| 462 | /* | ||
| 463 | * Lock I-TLB entry | ||
| 464 | * | ||
| 465 | * r0: Virtual address to translate and lock | ||
| 466 | */ | ||
| 467 | .align 5 | ||
| 468 | ENTRY(xscale_itlb_lock) | ||
| 469 | mrs r2, cpsr | ||
| 470 | orr r3, r2, #PSR_F_BIT | PSR_I_BIT | ||
| 471 | msr cpsr_c, r3 @ Disable interrupts | ||
| 472 | mcr p15, 0, r0, c8, c5, 1 @ Invalidate I-TLB entry | ||
| 473 | mcr p15, 0, r0, c10, c4, 0 @ Translate and lock | ||
| 474 | msr cpsr_c, r2 @ Restore interrupts | ||
| 475 | cpwait_ret lr, ip | ||
| 476 | |||
| 477 | /* | ||
| 478 | * Lock D-TLB entry | ||
| 479 | * | ||
| 480 | * r0: Virtual address to translate and lock | ||
| 481 | */ | ||
| 482 | .align 5 | ||
| 483 | ENTRY(xscale_dtlb_lock) | ||
| 484 | mrs r2, cpsr | ||
| 485 | orr r3, r2, #PSR_F_BIT | PSR_I_BIT | ||
| 486 | msr cpsr_c, r3 @ Disable interrupts | ||
| 487 | mcr p15, 0, r0, c8, c6, 1 @ Invalidate D-TLB entry | ||
| 488 | mcr p15, 0, r0, c10, c8, 0 @ Translate and lock | ||
| 489 | msr cpsr_c, r2 @ Restore interrupts | ||
| 490 | cpwait_ret lr, ip | ||
| 491 | |||
| 492 | /* | ||
| 493 | * Unlock all I-TLB entries | ||
| 494 | */ | ||
| 495 | .align 5 | ||
| 496 | ENTRY(xscale_itlb_unlock) | ||
| 497 | mcr p15, 0, ip, c10, c4, 1 @ Unlock I-TLB | ||
| 498 | mcr p15, 0, ip, c8, c5, 0 @ Invalidate I-TLB | ||
| 499 | cpwait_ret lr, ip | ||
| 500 | |||
| 501 | /* | ||
| 502 | * Unlock all D-TLB entries | ||
| 503 | */ | ||
| 504 | ENTRY(xscale_dtlb_unlock) | ||
| 505 | mcr p15, 0, ip, c10, c8, 1 @ Unlock D-TBL | ||
| 506 | mcr p15, 0, ip, c8, c6, 0 @ Invalidate D-TLB | ||
| 507 | cpwait_ret lr, ip | ||
| 508 | |||
| 509 | /* =============================== PageTable ============================== */ | 373 | /* =============================== PageTable ============================== */ |
| 510 | 374 | ||
| 511 | #define PTE_CACHE_WRITE_ALLOCATE 0 | 375 | #define PTE_CACHE_WRITE_ALLOCATE 0 |
diff --git a/arch/arm/nwfpe/double_cpdo.c b/arch/arm/nwfpe/double_cpdo.c index 7ffd8cb9bc96..c51d1386a97c 100644 --- a/arch/arm/nwfpe/double_cpdo.c +++ b/arch/arm/nwfpe/double_cpdo.c | |||
| @@ -40,17 +40,17 @@ float64 float64_arccos(float64 rFm); | |||
| 40 | float64 float64_pow(float64 rFn, float64 rFm); | 40 | float64 float64_pow(float64 rFn, float64 rFm); |
| 41 | float64 float64_pol(float64 rFn, float64 rFm); | 41 | float64 float64_pol(float64 rFn, float64 rFm); |
| 42 | 42 | ||
| 43 | static float64 float64_rsf(float64 rFn, float64 rFm) | 43 | static float64 float64_rsf(struct roundingData *roundData, float64 rFn, float64 rFm) |
| 44 | { | 44 | { |
| 45 | return float64_sub(rFm, rFn); | 45 | return float64_sub(roundData, rFm, rFn); |
| 46 | } | 46 | } |
| 47 | 47 | ||
| 48 | static float64 float64_rdv(float64 rFn, float64 rFm) | 48 | static float64 float64_rdv(struct roundingData *roundData, float64 rFn, float64 rFm) |
| 49 | { | 49 | { |
| 50 | return float64_div(rFm, rFn); | 50 | return float64_div(roundData, rFm, rFn); |
| 51 | } | 51 | } |
| 52 | 52 | ||
| 53 | static float64 (*const dyadic_double[16])(float64 rFn, float64 rFm) = { | 53 | static float64 (*const dyadic_double[16])(struct roundingData*, float64 rFn, float64 rFm) = { |
| 54 | [ADF_CODE >> 20] = float64_add, | 54 | [ADF_CODE >> 20] = float64_add, |
| 55 | [MUF_CODE >> 20] = float64_mul, | 55 | [MUF_CODE >> 20] = float64_mul, |
| 56 | [SUF_CODE >> 20] = float64_sub, | 56 | [SUF_CODE >> 20] = float64_sub, |
| @@ -65,12 +65,12 @@ static float64 (*const dyadic_double[16])(float64 rFn, float64 rFm) = { | |||
| 65 | [FRD_CODE >> 20] = float64_rdv, | 65 | [FRD_CODE >> 20] = float64_rdv, |
| 66 | }; | 66 | }; |
| 67 | 67 | ||
| 68 | static float64 float64_mvf(float64 rFm) | 68 | static float64 float64_mvf(struct roundingData *roundData,float64 rFm) |
| 69 | { | 69 | { |
| 70 | return rFm; | 70 | return rFm; |
| 71 | } | 71 | } |
| 72 | 72 | ||
| 73 | static float64 float64_mnf(float64 rFm) | 73 | static float64 float64_mnf(struct roundingData *roundData,float64 rFm) |
| 74 | { | 74 | { |
| 75 | union float64_components u; | 75 | union float64_components u; |
| 76 | 76 | ||
| @@ -84,7 +84,7 @@ static float64 float64_mnf(float64 rFm) | |||
| 84 | return u.f64; | 84 | return u.f64; |
| 85 | } | 85 | } |
| 86 | 86 | ||
| 87 | static float64 float64_abs(float64 rFm) | 87 | static float64 float64_abs(struct roundingData *roundData,float64 rFm) |
| 88 | { | 88 | { |
| 89 | union float64_components u; | 89 | union float64_components u; |
| 90 | 90 | ||
| @@ -98,7 +98,7 @@ static float64 float64_abs(float64 rFm) | |||
| 98 | return u.f64; | 98 | return u.f64; |
| 99 | } | 99 | } |
| 100 | 100 | ||
| 101 | static float64 (*const monadic_double[16])(float64 rFm) = { | 101 | static float64 (*const monadic_double[16])(struct roundingData *, float64 rFm) = { |
| 102 | [MVF_CODE >> 20] = float64_mvf, | 102 | [MVF_CODE >> 20] = float64_mvf, |
| 103 | [MNF_CODE >> 20] = float64_mnf, | 103 | [MNF_CODE >> 20] = float64_mnf, |
| 104 | [ABS_CODE >> 20] = float64_abs, | 104 | [ABS_CODE >> 20] = float64_abs, |
| @@ -108,7 +108,7 @@ static float64 (*const monadic_double[16])(float64 rFm) = { | |||
| 108 | [NRM_CODE >> 20] = float64_mvf, | 108 | [NRM_CODE >> 20] = float64_mvf, |
| 109 | }; | 109 | }; |
| 110 | 110 | ||
| 111 | unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd) | 111 | unsigned int DoubleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd) |
| 112 | { | 112 | { |
| 113 | FPA11 *fpa11 = GET_FPA11(); | 113 | FPA11 *fpa11 = GET_FPA11(); |
| 114 | float64 rFm; | 114 | float64 rFm; |
| @@ -151,13 +151,13 @@ unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd) | |||
| 151 | } | 151 | } |
| 152 | 152 | ||
| 153 | if (dyadic_double[opc_mask_shift]) { | 153 | if (dyadic_double[opc_mask_shift]) { |
| 154 | rFd->fDouble = dyadic_double[opc_mask_shift](rFn, rFm); | 154 | rFd->fDouble = dyadic_double[opc_mask_shift](roundData, rFn, rFm); |
| 155 | } else { | 155 | } else { |
| 156 | return 0; | 156 | return 0; |
| 157 | } | 157 | } |
| 158 | } else { | 158 | } else { |
| 159 | if (monadic_double[opc_mask_shift]) { | 159 | if (monadic_double[opc_mask_shift]) { |
| 160 | rFd->fDouble = monadic_double[opc_mask_shift](rFm); | 160 | rFd->fDouble = monadic_double[opc_mask_shift](roundData, rFm); |
| 161 | } else { | 161 | } else { |
| 162 | return 0; | 162 | return 0; |
| 163 | } | 163 | } |
diff --git a/arch/arm/nwfpe/extended_cpdo.c b/arch/arm/nwfpe/extended_cpdo.c index c39f68a3449e..65a279ba927f 100644 --- a/arch/arm/nwfpe/extended_cpdo.c +++ b/arch/arm/nwfpe/extended_cpdo.c | |||
| @@ -35,17 +35,17 @@ floatx80 floatx80_arccos(floatx80 rFm); | |||
| 35 | floatx80 floatx80_pow(floatx80 rFn, floatx80 rFm); | 35 | floatx80 floatx80_pow(floatx80 rFn, floatx80 rFm); |
| 36 | floatx80 floatx80_pol(floatx80 rFn, floatx80 rFm); | 36 | floatx80 floatx80_pol(floatx80 rFn, floatx80 rFm); |
| 37 | 37 | ||
| 38 | static floatx80 floatx80_rsf(floatx80 rFn, floatx80 rFm) | 38 | static floatx80 floatx80_rsf(struct roundingData *roundData, floatx80 rFn, floatx80 rFm) |
| 39 | { | 39 | { |
| 40 | return floatx80_sub(rFm, rFn); | 40 | return floatx80_sub(roundData, rFm, rFn); |
| 41 | } | 41 | } |
| 42 | 42 | ||
| 43 | static floatx80 floatx80_rdv(floatx80 rFn, floatx80 rFm) | 43 | static floatx80 floatx80_rdv(struct roundingData *roundData, floatx80 rFn, floatx80 rFm) |
| 44 | { | 44 | { |
| 45 | return floatx80_div(rFm, rFn); | 45 | return floatx80_div(roundData, rFm, rFn); |
| 46 | } | 46 | } |
| 47 | 47 | ||
| 48 | static floatx80 (*const dyadic_extended[16])(floatx80 rFn, floatx80 rFm) = { | 48 | static floatx80 (*const dyadic_extended[16])(struct roundingData*, floatx80 rFn, floatx80 rFm) = { |
| 49 | [ADF_CODE >> 20] = floatx80_add, | 49 | [ADF_CODE >> 20] = floatx80_add, |
| 50 | [MUF_CODE >> 20] = floatx80_mul, | 50 | [MUF_CODE >> 20] = floatx80_mul, |
| 51 | [SUF_CODE >> 20] = floatx80_sub, | 51 | [SUF_CODE >> 20] = floatx80_sub, |
| @@ -60,24 +60,24 @@ static floatx80 (*const dyadic_extended[16])(floatx80 rFn, floatx80 rFm) = { | |||
| 60 | [FRD_CODE >> 20] = floatx80_rdv, | 60 | [FRD_CODE >> 20] = floatx80_rdv, |
| 61 | }; | 61 | }; |
| 62 | 62 | ||
| 63 | static floatx80 floatx80_mvf(floatx80 rFm) | 63 | static floatx80 floatx80_mvf(struct roundingData *roundData, floatx80 rFm) |
| 64 | { | 64 | { |
| 65 | return rFm; | 65 | return rFm; |
| 66 | } | 66 | } |
| 67 | 67 | ||
| 68 | static floatx80 floatx80_mnf(floatx80 rFm) | 68 | static floatx80 floatx80_mnf(struct roundingData *roundData, floatx80 rFm) |
| 69 | { | 69 | { |
| 70 | rFm.high ^= 0x8000; | 70 | rFm.high ^= 0x8000; |
| 71 | return rFm; | 71 | return rFm; |
| 72 | } | 72 | } |
| 73 | 73 | ||
| 74 | static floatx80 floatx80_abs(floatx80 rFm) | 74 | static floatx80 floatx80_abs(struct roundingData *roundData, floatx80 rFm) |
| 75 | { | 75 | { |
| 76 | rFm.high &= 0x7fff; | 76 | rFm.high &= 0x7fff; |
| 77 | return rFm; | 77 | return rFm; |
| 78 | } | 78 | } |
| 79 | 79 | ||
| 80 | static floatx80 (*const monadic_extended[16])(floatx80 rFm) = { | 80 | static floatx80 (*const monadic_extended[16])(struct roundingData*, floatx80 rFm) = { |
| 81 | [MVF_CODE >> 20] = floatx80_mvf, | 81 | [MVF_CODE >> 20] = floatx80_mvf, |
| 82 | [MNF_CODE >> 20] = floatx80_mnf, | 82 | [MNF_CODE >> 20] = floatx80_mnf, |
| 83 | [ABS_CODE >> 20] = floatx80_abs, | 83 | [ABS_CODE >> 20] = floatx80_abs, |
| @@ -87,7 +87,7 @@ static floatx80 (*const monadic_extended[16])(floatx80 rFm) = { | |||
| 87 | [NRM_CODE >> 20] = floatx80_mvf, | 87 | [NRM_CODE >> 20] = floatx80_mvf, |
| 88 | }; | 88 | }; |
| 89 | 89 | ||
| 90 | unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd) | 90 | unsigned int ExtendedCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd) |
| 91 | { | 91 | { |
| 92 | FPA11 *fpa11 = GET_FPA11(); | 92 | FPA11 *fpa11 = GET_FPA11(); |
| 93 | floatx80 rFm; | 93 | floatx80 rFm; |
| @@ -138,13 +138,13 @@ unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd) | |||
| 138 | } | 138 | } |
| 139 | 139 | ||
| 140 | if (dyadic_extended[opc_mask_shift]) { | 140 | if (dyadic_extended[opc_mask_shift]) { |
| 141 | rFd->fExtended = dyadic_extended[opc_mask_shift](rFn, rFm); | 141 | rFd->fExtended = dyadic_extended[opc_mask_shift](roundData, rFn, rFm); |
| 142 | } else { | 142 | } else { |
| 143 | return 0; | 143 | return 0; |
| 144 | } | 144 | } |
| 145 | } else { | 145 | } else { |
| 146 | if (monadic_extended[opc_mask_shift]) { | 146 | if (monadic_extended[opc_mask_shift]) { |
| 147 | rFd->fExtended = monadic_extended[opc_mask_shift](rFm); | 147 | rFd->fExtended = monadic_extended[opc_mask_shift](roundData, rFm); |
| 148 | } else { | 148 | } else { |
| 149 | return 0; | 149 | return 0; |
| 150 | } | 150 | } |
diff --git a/arch/arm/nwfpe/fpa11.c b/arch/arm/nwfpe/fpa11.c index bf61696865ec..7690f731ee87 100644 --- a/arch/arm/nwfpe/fpa11.c +++ b/arch/arm/nwfpe/fpa11.c | |||
| @@ -51,48 +51,42 @@ static void resetFPA11(void) | |||
| 51 | fpa11->fpsr = FP_EMULATOR | BIT_AC; | 51 | fpa11->fpsr = FP_EMULATOR | BIT_AC; |
| 52 | } | 52 | } |
| 53 | 53 | ||
| 54 | void SetRoundingMode(const unsigned int opcode) | 54 | int8 SetRoundingMode(const unsigned int opcode) |
| 55 | { | 55 | { |
| 56 | switch (opcode & MASK_ROUNDING_MODE) { | 56 | switch (opcode & MASK_ROUNDING_MODE) { |
| 57 | default: | 57 | default: |
| 58 | case ROUND_TO_NEAREST: | 58 | case ROUND_TO_NEAREST: |
| 59 | float_rounding_mode = float_round_nearest_even; | 59 | return float_round_nearest_even; |
| 60 | break; | ||
| 61 | 60 | ||
| 62 | case ROUND_TO_PLUS_INFINITY: | 61 | case ROUND_TO_PLUS_INFINITY: |
| 63 | float_rounding_mode = float_round_up; | 62 | return float_round_up; |
| 64 | break; | ||
| 65 | 63 | ||
| 66 | case ROUND_TO_MINUS_INFINITY: | 64 | case ROUND_TO_MINUS_INFINITY: |
| 67 | float_rounding_mode = float_round_down; | 65 | return float_round_down; |
| 68 | break; | ||
| 69 | 66 | ||
| 70 | case ROUND_TO_ZERO: | 67 | case ROUND_TO_ZERO: |
| 71 | float_rounding_mode = float_round_to_zero; | 68 | return float_round_to_zero; |
| 72 | break; | ||
| 73 | } | 69 | } |
| 74 | } | 70 | } |
| 75 | 71 | ||
| 76 | void SetRoundingPrecision(const unsigned int opcode) | 72 | int8 SetRoundingPrecision(const unsigned int opcode) |
| 77 | { | 73 | { |
| 78 | #ifdef CONFIG_FPE_NWFPE_XP | 74 | #ifdef CONFIG_FPE_NWFPE_XP |
| 79 | switch (opcode & MASK_ROUNDING_PRECISION) { | 75 | switch (opcode & MASK_ROUNDING_PRECISION) { |
| 80 | case ROUND_SINGLE: | 76 | case ROUND_SINGLE: |
| 81 | floatx80_rounding_precision = 32; | 77 | return 32; |
| 82 | break; | ||
| 83 | 78 | ||
| 84 | case ROUND_DOUBLE: | 79 | case ROUND_DOUBLE: |
| 85 | floatx80_rounding_precision = 64; | 80 | return 64; |
| 86 | break; | ||
| 87 | 81 | ||
| 88 | case ROUND_EXTENDED: | 82 | case ROUND_EXTENDED: |
| 89 | floatx80_rounding_precision = 80; | 83 | return 80; |
| 90 | break; | ||
| 91 | 84 | ||
| 92 | default: | 85 | default: |
| 93 | floatx80_rounding_precision = 80; | 86 | return 80; |
| 94 | } | 87 | } |
| 95 | #endif | 88 | #endif |
| 89 | return 80; | ||
| 96 | } | 90 | } |
| 97 | 91 | ||
| 98 | void nwfpe_init_fpa(union fp_state *fp) | 92 | void nwfpe_init_fpa(union fp_state *fp) |
| @@ -103,8 +97,6 @@ void nwfpe_init_fpa(union fp_state *fp) | |||
| 103 | #endif | 97 | #endif |
| 104 | memset(fpa11, 0, sizeof(FPA11)); | 98 | memset(fpa11, 0, sizeof(FPA11)); |
| 105 | resetFPA11(); | 99 | resetFPA11(); |
| 106 | SetRoundingMode(ROUND_TO_NEAREST); | ||
| 107 | SetRoundingPrecision(ROUND_EXTENDED); | ||
| 108 | fpa11->initflag = 1; | 100 | fpa11->initflag = 1; |
| 109 | } | 101 | } |
| 110 | 102 | ||
diff --git a/arch/arm/nwfpe/fpa11.h b/arch/arm/nwfpe/fpa11.h index e4a61aea534b..93523ae4b7a1 100644 --- a/arch/arm/nwfpe/fpa11.h +++ b/arch/arm/nwfpe/fpa11.h | |||
| @@ -37,6 +37,13 @@ | |||
| 37 | /* includes */ | 37 | /* includes */ |
| 38 | #include "fpsr.h" /* FP control and status register definitions */ | 38 | #include "fpsr.h" /* FP control and status register definitions */ |
| 39 | #include "milieu.h" | 39 | #include "milieu.h" |
| 40 | |||
| 41 | struct roundingData { | ||
| 42 | int8 mode; | ||
| 43 | int8 precision; | ||
| 44 | signed char exception; | ||
| 45 | }; | ||
| 46 | |||
| 40 | #include "softfloat.h" | 47 | #include "softfloat.h" |
| 41 | 48 | ||
| 42 | #define typeNone 0x00 | 49 | #define typeNone 0x00 |
| @@ -84,8 +91,8 @@ typedef struct tagFPA11 { | |||
| 84 | initialised. */ | 91 | initialised. */ |
| 85 | } FPA11; | 92 | } FPA11; |
| 86 | 93 | ||
| 87 | extern void SetRoundingMode(const unsigned int); | 94 | extern int8 SetRoundingMode(const unsigned int); |
| 88 | extern void SetRoundingPrecision(const unsigned int); | 95 | extern int8 SetRoundingPrecision(const unsigned int); |
| 89 | extern void nwfpe_init_fpa(union fp_state *fp); | 96 | extern void nwfpe_init_fpa(union fp_state *fp); |
| 90 | 97 | ||
| 91 | #endif | 98 | #endif |
diff --git a/arch/arm/nwfpe/fpa11_cpdo.c b/arch/arm/nwfpe/fpa11_cpdo.c index 1bea67437b6f..4a31dfd94068 100644 --- a/arch/arm/nwfpe/fpa11_cpdo.c +++ b/arch/arm/nwfpe/fpa11_cpdo.c | |||
| @@ -24,15 +24,16 @@ | |||
| 24 | #include "fpa11.h" | 24 | #include "fpa11.h" |
| 25 | #include "fpopcode.h" | 25 | #include "fpopcode.h" |
| 26 | 26 | ||
| 27 | unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd); | 27 | unsigned int SingleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd); |
| 28 | unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd); | 28 | unsigned int DoubleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd); |
| 29 | unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd); | 29 | unsigned int ExtendedCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd); |
| 30 | 30 | ||
| 31 | unsigned int EmulateCPDO(const unsigned int opcode) | 31 | unsigned int EmulateCPDO(const unsigned int opcode) |
| 32 | { | 32 | { |
| 33 | FPA11 *fpa11 = GET_FPA11(); | 33 | FPA11 *fpa11 = GET_FPA11(); |
| 34 | FPREG *rFd; | 34 | FPREG *rFd; |
| 35 | unsigned int nType, nDest, nRc; | 35 | unsigned int nType, nDest, nRc; |
| 36 | struct roundingData roundData; | ||
| 36 | 37 | ||
| 37 | /* Get the destination size. If not valid let Linux perform | 38 | /* Get the destination size. If not valid let Linux perform |
| 38 | an invalid instruction trap. */ | 39 | an invalid instruction trap. */ |
| @@ -40,7 +41,9 @@ unsigned int EmulateCPDO(const unsigned int opcode) | |||
| 40 | if (typeNone == nDest) | 41 | if (typeNone == nDest) |
| 41 | return 0; | 42 | return 0; |
| 42 | 43 | ||
| 43 | SetRoundingMode(opcode); | 44 | roundData.mode = SetRoundingMode(opcode); |
| 45 | roundData.precision = SetRoundingPrecision(opcode); | ||
| 46 | roundData.exception = 0; | ||
| 44 | 47 | ||
| 45 | /* Compare the size of the operands in Fn and Fm. | 48 | /* Compare the size of the operands in Fn and Fm. |
| 46 | Choose the largest size and perform operations in that size, | 49 | Choose the largest size and perform operations in that size, |
| @@ -63,14 +66,14 @@ unsigned int EmulateCPDO(const unsigned int opcode) | |||
| 63 | 66 | ||
| 64 | switch (nType) { | 67 | switch (nType) { |
| 65 | case typeSingle: | 68 | case typeSingle: |
| 66 | nRc = SingleCPDO(opcode, rFd); | 69 | nRc = SingleCPDO(&roundData, opcode, rFd); |
| 67 | break; | 70 | break; |
| 68 | case typeDouble: | 71 | case typeDouble: |
| 69 | nRc = DoubleCPDO(opcode, rFd); | 72 | nRc = DoubleCPDO(&roundData, opcode, rFd); |
| 70 | break; | 73 | break; |
| 71 | #ifdef CONFIG_FPE_NWFPE_XP | 74 | #ifdef CONFIG_FPE_NWFPE_XP |
| 72 | case typeExtended: | 75 | case typeExtended: |
| 73 | nRc = ExtendedCPDO(opcode, rFd); | 76 | nRc = ExtendedCPDO(&roundData, opcode, rFd); |
| 74 | break; | 77 | break; |
| 75 | #endif | 78 | #endif |
| 76 | default: | 79 | default: |
| @@ -93,9 +96,9 @@ unsigned int EmulateCPDO(const unsigned int opcode) | |||
| 93 | case typeSingle: | 96 | case typeSingle: |
| 94 | { | 97 | { |
| 95 | if (typeDouble == nType) | 98 | if (typeDouble == nType) |
| 96 | rFd->fSingle = float64_to_float32(rFd->fDouble); | 99 | rFd->fSingle = float64_to_float32(&roundData, rFd->fDouble); |
| 97 | else | 100 | else |
| 98 | rFd->fSingle = floatx80_to_float32(rFd->fExtended); | 101 | rFd->fSingle = floatx80_to_float32(&roundData, rFd->fExtended); |
| 99 | } | 102 | } |
| 100 | break; | 103 | break; |
| 101 | 104 | ||
| @@ -104,7 +107,7 @@ unsigned int EmulateCPDO(const unsigned int opcode) | |||
| 104 | if (typeSingle == nType) | 107 | if (typeSingle == nType) |
| 105 | rFd->fDouble = float32_to_float64(rFd->fSingle); | 108 | rFd->fDouble = float32_to_float64(rFd->fSingle); |
| 106 | else | 109 | else |
| 107 | rFd->fDouble = floatx80_to_float64(rFd->fExtended); | 110 | rFd->fDouble = floatx80_to_float64(&roundData, rFd->fExtended); |
| 108 | } | 111 | } |
| 109 | break; | 112 | break; |
| 110 | 113 | ||
| @@ -121,12 +124,15 @@ unsigned int EmulateCPDO(const unsigned int opcode) | |||
| 121 | #else | 124 | #else |
| 122 | if (nDest != nType) { | 125 | if (nDest != nType) { |
| 123 | if (nDest == typeSingle) | 126 | if (nDest == typeSingle) |
| 124 | rFd->fSingle = float64_to_float32(rFd->fDouble); | 127 | rFd->fSingle = float64_to_float32(&roundData, rFd->fDouble); |
| 125 | else | 128 | else |
| 126 | rFd->fDouble = float32_to_float64(rFd->fSingle); | 129 | rFd->fDouble = float32_to_float64(rFd->fSingle); |
| 127 | } | 130 | } |
| 128 | #endif | 131 | #endif |
| 129 | } | 132 | } |
| 130 | 133 | ||
| 134 | if (roundData.exception) | ||
| 135 | float_raise(roundData.exception); | ||
| 136 | |||
| 131 | return nRc; | 137 | return nRc; |
| 132 | } | 138 | } |
diff --git a/arch/arm/nwfpe/fpa11_cpdt.c b/arch/arm/nwfpe/fpa11_cpdt.c index 95fb63fa9d18..b0db5cbcc3b1 100644 --- a/arch/arm/nwfpe/fpa11_cpdt.c +++ b/arch/arm/nwfpe/fpa11_cpdt.c | |||
| @@ -96,7 +96,7 @@ static inline void loadMultiple(const unsigned int Fn, const unsigned int __user | |||
| 96 | } | 96 | } |
| 97 | } | 97 | } |
| 98 | 98 | ||
| 99 | static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem) | 99 | static inline void storeSingle(struct roundingData *roundData, const unsigned int Fn, unsigned int __user *pMem) |
| 100 | { | 100 | { |
| 101 | FPA11 *fpa11 = GET_FPA11(); | 101 | FPA11 *fpa11 = GET_FPA11(); |
| 102 | union { | 102 | union { |
| @@ -106,12 +106,12 @@ static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem) | |||
| 106 | 106 | ||
| 107 | switch (fpa11->fType[Fn]) { | 107 | switch (fpa11->fType[Fn]) { |
| 108 | case typeDouble: | 108 | case typeDouble: |
| 109 | val.f = float64_to_float32(fpa11->fpreg[Fn].fDouble); | 109 | val.f = float64_to_float32(roundData, fpa11->fpreg[Fn].fDouble); |
| 110 | break; | 110 | break; |
| 111 | 111 | ||
| 112 | #ifdef CONFIG_FPE_NWFPE_XP | 112 | #ifdef CONFIG_FPE_NWFPE_XP |
| 113 | case typeExtended: | 113 | case typeExtended: |
| 114 | val.f = floatx80_to_float32(fpa11->fpreg[Fn].fExtended); | 114 | val.f = floatx80_to_float32(roundData, fpa11->fpreg[Fn].fExtended); |
| 115 | break; | 115 | break; |
| 116 | #endif | 116 | #endif |
| 117 | 117 | ||
| @@ -122,7 +122,7 @@ static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem) | |||
| 122 | put_user(val.i[0], pMem); | 122 | put_user(val.i[0], pMem); |
| 123 | } | 123 | } |
| 124 | 124 | ||
| 125 | static inline void storeDouble(const unsigned int Fn, unsigned int __user *pMem) | 125 | static inline void storeDouble(struct roundingData *roundData, const unsigned int Fn, unsigned int __user *pMem) |
| 126 | { | 126 | { |
| 127 | FPA11 *fpa11 = GET_FPA11(); | 127 | FPA11 *fpa11 = GET_FPA11(); |
| 128 | union { | 128 | union { |
| @@ -137,7 +137,7 @@ static inline void storeDouble(const unsigned int Fn, unsigned int __user *pMem) | |||
| 137 | 137 | ||
| 138 | #ifdef CONFIG_FPE_NWFPE_XP | 138 | #ifdef CONFIG_FPE_NWFPE_XP |
| 139 | case typeExtended: | 139 | case typeExtended: |
| 140 | val.f = floatx80_to_float64(fpa11->fpreg[Fn].fExtended); | 140 | val.f = floatx80_to_float64(roundData, fpa11->fpreg[Fn].fExtended); |
| 141 | break; | 141 | break; |
| 142 | #endif | 142 | #endif |
| 143 | 143 | ||
| @@ -259,8 +259,11 @@ unsigned int PerformSTF(const unsigned int opcode) | |||
| 259 | { | 259 | { |
| 260 | unsigned int __user *pBase, *pAddress, *pFinal; | 260 | unsigned int __user *pBase, *pAddress, *pFinal; |
| 261 | unsigned int nRc = 1, write_back = WRITE_BACK(opcode); | 261 | unsigned int nRc = 1, write_back = WRITE_BACK(opcode); |
| 262 | struct roundingData roundData; | ||
| 262 | 263 | ||
| 263 | SetRoundingMode(ROUND_TO_NEAREST); | 264 | roundData.mode = SetRoundingMode(opcode); |
| 265 | roundData.precision = SetRoundingPrecision(opcode); | ||
| 266 | roundData.exception = 0; | ||
| 264 | 267 | ||
| 265 | pBase = (unsigned int __user *) readRegister(getRn(opcode)); | 268 | pBase = (unsigned int __user *) readRegister(getRn(opcode)); |
| 266 | if (REG_PC == getRn(opcode)) { | 269 | if (REG_PC == getRn(opcode)) { |
| @@ -281,10 +284,10 @@ unsigned int PerformSTF(const unsigned int opcode) | |||
| 281 | 284 | ||
| 282 | switch (opcode & MASK_TRANSFER_LENGTH) { | 285 | switch (opcode & MASK_TRANSFER_LENGTH) { |
| 283 | case TRANSFER_SINGLE: | 286 | case TRANSFER_SINGLE: |
| 284 | storeSingle(getFd(opcode), pAddress); | 287 | storeSingle(&roundData, getFd(opcode), pAddress); |
| 285 | break; | 288 | break; |
| 286 | case TRANSFER_DOUBLE: | 289 | case TRANSFER_DOUBLE: |
| 287 | storeDouble(getFd(opcode), pAddress); | 290 | storeDouble(&roundData, getFd(opcode), pAddress); |
| 288 | break; | 291 | break; |
| 289 | #ifdef CONFIG_FPE_NWFPE_XP | 292 | #ifdef CONFIG_FPE_NWFPE_XP |
| 290 | case TRANSFER_EXTENDED: | 293 | case TRANSFER_EXTENDED: |
| @@ -295,6 +298,9 @@ unsigned int PerformSTF(const unsigned int opcode) | |||
| 295 | nRc = 0; | 298 | nRc = 0; |
| 296 | } | 299 | } |
| 297 | 300 | ||
| 301 | if (roundData.exception) | ||
| 302 | float_raise(roundData.exception); | ||
| 303 | |||
| 298 | if (write_back) | 304 | if (write_back) |
| 299 | writeRegister(getRn(opcode), (unsigned long) pFinal); | 305 | writeRegister(getRn(opcode), (unsigned long) pFinal); |
| 300 | return nRc; | 306 | return nRc; |
diff --git a/arch/arm/nwfpe/fpa11_cprt.c b/arch/arm/nwfpe/fpa11_cprt.c index db01fbc97216..adf8d3000540 100644 --- a/arch/arm/nwfpe/fpa11_cprt.c +++ b/arch/arm/nwfpe/fpa11_cprt.c | |||
| @@ -33,8 +33,6 @@ extern flag floatx80_is_nan(floatx80); | |||
| 33 | extern flag float64_is_nan(float64); | 33 | extern flag float64_is_nan(float64); |
| 34 | extern flag float32_is_nan(float32); | 34 | extern flag float32_is_nan(float32); |
| 35 | 35 | ||
| 36 | void SetRoundingMode(const unsigned int opcode); | ||
| 37 | |||
| 38 | unsigned int PerformFLT(const unsigned int opcode); | 36 | unsigned int PerformFLT(const unsigned int opcode); |
| 39 | unsigned int PerformFIX(const unsigned int opcode); | 37 | unsigned int PerformFIX(const unsigned int opcode); |
| 40 | 38 | ||
| @@ -77,14 +75,17 @@ unsigned int EmulateCPRT(const unsigned int opcode) | |||
| 77 | unsigned int PerformFLT(const unsigned int opcode) | 75 | unsigned int PerformFLT(const unsigned int opcode) |
| 78 | { | 76 | { |
| 79 | FPA11 *fpa11 = GET_FPA11(); | 77 | FPA11 *fpa11 = GET_FPA11(); |
| 80 | SetRoundingMode(opcode); | 78 | struct roundingData roundData; |
| 81 | SetRoundingPrecision(opcode); | 79 | |
| 80 | roundData.mode = SetRoundingMode(opcode); | ||
| 81 | roundData.precision = SetRoundingPrecision(opcode); | ||
| 82 | roundData.exception = 0; | ||
| 82 | 83 | ||
| 83 | switch (opcode & MASK_ROUNDING_PRECISION) { | 84 | switch (opcode & MASK_ROUNDING_PRECISION) { |
| 84 | case ROUND_SINGLE: | 85 | case ROUND_SINGLE: |
| 85 | { | 86 | { |
| 86 | fpa11->fType[getFn(opcode)] = typeSingle; | 87 | fpa11->fType[getFn(opcode)] = typeSingle; |
| 87 | fpa11->fpreg[getFn(opcode)].fSingle = int32_to_float32(readRegister(getRd(opcode))); | 88 | fpa11->fpreg[getFn(opcode)].fSingle = int32_to_float32(&roundData, readRegister(getRd(opcode))); |
| 88 | } | 89 | } |
| 89 | break; | 90 | break; |
| 90 | 91 | ||
| @@ -108,6 +109,9 @@ unsigned int PerformFLT(const unsigned int opcode) | |||
| 108 | return 0; | 109 | return 0; |
| 109 | } | 110 | } |
| 110 | 111 | ||
| 112 | if (roundData.exception) | ||
| 113 | float_raise(roundData.exception); | ||
| 114 | |||
| 111 | return 1; | 115 | return 1; |
| 112 | } | 116 | } |
| 113 | 117 | ||
| @@ -115,26 +119,29 @@ unsigned int PerformFIX(const unsigned int opcode) | |||
| 115 | { | 119 | { |
| 116 | FPA11 *fpa11 = GET_FPA11(); | 120 | FPA11 *fpa11 = GET_FPA11(); |
| 117 | unsigned int Fn = getFm(opcode); | 121 | unsigned int Fn = getFm(opcode); |
| 122 | struct roundingData roundData; | ||
| 118 | 123 | ||
| 119 | SetRoundingMode(opcode); | 124 | roundData.mode = SetRoundingMode(opcode); |
| 125 | roundData.precision = SetRoundingPrecision(opcode); | ||
| 126 | roundData.exception = 0; | ||
| 120 | 127 | ||
| 121 | switch (fpa11->fType[Fn]) { | 128 | switch (fpa11->fType[Fn]) { |
| 122 | case typeSingle: | 129 | case typeSingle: |
| 123 | { | 130 | { |
| 124 | writeRegister(getRd(opcode), float32_to_int32(fpa11->fpreg[Fn].fSingle)); | 131 | writeRegister(getRd(opcode), float32_to_int32(&roundData, fpa11->fpreg[Fn].fSingle)); |
| 125 | } | 132 | } |
| 126 | break; | 133 | break; |
| 127 | 134 | ||
| 128 | case typeDouble: | 135 | case typeDouble: |
| 129 | { | 136 | { |
| 130 | writeRegister(getRd(opcode), float64_to_int32(fpa11->fpreg[Fn].fDouble)); | 137 | writeRegister(getRd(opcode), float64_to_int32(&roundData, fpa11->fpreg[Fn].fDouble)); |
| 131 | } | 138 | } |
| 132 | break; | 139 | break; |
| 133 | 140 | ||
| 134 | #ifdef CONFIG_FPE_NWFPE_XP | 141 | #ifdef CONFIG_FPE_NWFPE_XP |
| 135 | case typeExtended: | 142 | case typeExtended: |
| 136 | { | 143 | { |
| 137 | writeRegister(getRd(opcode), floatx80_to_int32(fpa11->fpreg[Fn].fExtended)); | 144 | writeRegister(getRd(opcode), floatx80_to_int32(&roundData, fpa11->fpreg[Fn].fExtended)); |
| 138 | } | 145 | } |
| 139 | break; | 146 | break; |
| 140 | #endif | 147 | #endif |
| @@ -143,6 +150,9 @@ unsigned int PerformFIX(const unsigned int opcode) | |||
| 143 | return 0; | 150 | return 0; |
| 144 | } | 151 | } |
| 145 | 152 | ||
| 153 | if (roundData.exception) | ||
| 154 | float_raise(roundData.exception); | ||
| 155 | |||
| 146 | return 1; | 156 | return 1; |
| 147 | } | 157 | } |
| 148 | 158 | ||
diff --git a/arch/arm/nwfpe/fpmodule.c b/arch/arm/nwfpe/fpmodule.c index 12885f31d347..2dfe1ac42ee8 100644 --- a/arch/arm/nwfpe/fpmodule.c +++ b/arch/arm/nwfpe/fpmodule.c | |||
| @@ -116,8 +116,6 @@ fpmodule.c to integrate with the NetBSD kernel (I hope!). | |||
| 116 | code to access data in user space in some other source files at the | 116 | code to access data in user space in some other source files at the |
| 117 | moment (grep for get_user / put_user calls). --philb] | 117 | moment (grep for get_user / put_user calls). --philb] |
| 118 | 118 | ||
| 119 | float_exception_flags is a global variable in SoftFloat. | ||
| 120 | |||
| 121 | This function is called by the SoftFloat routines to raise a floating | 119 | This function is called by the SoftFloat routines to raise a floating |
| 122 | point exception. We check the trap enable byte in the FPSR, and raise | 120 | point exception. We check the trap enable byte in the FPSR, and raise |
| 123 | a SIGFPE exception if necessary. If not the relevant bits in the | 121 | a SIGFPE exception if necessary. If not the relevant bits in the |
| @@ -129,15 +127,14 @@ void float_raise(signed char flags) | |||
| 129 | register unsigned int fpsr, cumulativeTraps; | 127 | register unsigned int fpsr, cumulativeTraps; |
| 130 | 128 | ||
| 131 | #ifdef CONFIG_DEBUG_USER | 129 | #ifdef CONFIG_DEBUG_USER |
| 132 | printk(KERN_DEBUG | 130 | /* Ignore inexact errors as there are far too many of them to log */ |
| 133 | "NWFPE: %s[%d] takes exception %08x at %p from %08lx\n", | 131 | if (flags & ~BIT_IXC) |
| 134 | current->comm, current->pid, flags, | 132 | printk(KERN_DEBUG |
| 135 | __builtin_return_address(0), GET_USERREG()->ARM_pc); | 133 | "NWFPE: %s[%d] takes exception %08x at %p from %08lx\n", |
| 134 | current->comm, current->pid, flags, | ||
| 135 | __builtin_return_address(0), GET_USERREG()->ARM_pc); | ||
| 136 | #endif | 136 | #endif |
| 137 | 137 | ||
| 138 | /* Keep SoftFloat exception flags up to date. */ | ||
| 139 | float_exception_flags |= flags; | ||
| 140 | |||
| 141 | /* Read fpsr and initialize the cumulativeTraps. */ | 138 | /* Read fpsr and initialize the cumulativeTraps. */ |
| 142 | fpsr = readFPSR(); | 139 | fpsr = readFPSR(); |
| 143 | cumulativeTraps = 0; | 140 | cumulativeTraps = 0; |
diff --git a/arch/arm/nwfpe/single_cpdo.c b/arch/arm/nwfpe/single_cpdo.c index 705808e88d9d..c66981d682cf 100644 --- a/arch/arm/nwfpe/single_cpdo.c +++ b/arch/arm/nwfpe/single_cpdo.c | |||
| @@ -36,17 +36,17 @@ float32 float32_arccos(float32 rFm); | |||
| 36 | float32 float32_pow(float32 rFn, float32 rFm); | 36 | float32 float32_pow(float32 rFn, float32 rFm); |
| 37 | float32 float32_pol(float32 rFn, float32 rFm); | 37 | float32 float32_pol(float32 rFn, float32 rFm); |
| 38 | 38 | ||
| 39 | static float32 float32_rsf(float32 rFn, float32 rFm) | 39 | static float32 float32_rsf(struct roundingData *roundData, float32 rFn, float32 rFm) |
| 40 | { | 40 | { |
| 41 | return float32_sub(rFm, rFn); | 41 | return float32_sub(roundData, rFm, rFn); |
| 42 | } | 42 | } |
| 43 | 43 | ||
| 44 | static float32 float32_rdv(float32 rFn, float32 rFm) | 44 | static float32 float32_rdv(struct roundingData *roundData, float32 rFn, float32 rFm) |
| 45 | { | 45 | { |
| 46 | return float32_div(rFm, rFn); | 46 | return float32_div(roundData, rFm, rFn); |
| 47 | } | 47 | } |
| 48 | 48 | ||
| 49 | static float32 (*const dyadic_single[16])(float32 rFn, float32 rFm) = { | 49 | static float32 (*const dyadic_single[16])(struct roundingData *, float32 rFn, float32 rFm) = { |
| 50 | [ADF_CODE >> 20] = float32_add, | 50 | [ADF_CODE >> 20] = float32_add, |
| 51 | [MUF_CODE >> 20] = float32_mul, | 51 | [MUF_CODE >> 20] = float32_mul, |
| 52 | [SUF_CODE >> 20] = float32_sub, | 52 | [SUF_CODE >> 20] = float32_sub, |
| @@ -60,22 +60,22 @@ static float32 (*const dyadic_single[16])(float32 rFn, float32 rFm) = { | |||
| 60 | [FRD_CODE >> 20] = float32_rdv, | 60 | [FRD_CODE >> 20] = float32_rdv, |
| 61 | }; | 61 | }; |
| 62 | 62 | ||
| 63 | static float32 float32_mvf(float32 rFm) | 63 | static float32 float32_mvf(struct roundingData *roundData, float32 rFm) |
| 64 | { | 64 | { |
| 65 | return rFm; | 65 | return rFm; |
| 66 | } | 66 | } |
| 67 | 67 | ||
| 68 | static float32 float32_mnf(float32 rFm) | 68 | static float32 float32_mnf(struct roundingData *roundData, float32 rFm) |
| 69 | { | 69 | { |
| 70 | return rFm ^ 0x80000000; | 70 | return rFm ^ 0x80000000; |
| 71 | } | 71 | } |
| 72 | 72 | ||
| 73 | static float32 float32_abs(float32 rFm) | 73 | static float32 float32_abs(struct roundingData *roundData, float32 rFm) |
| 74 | { | 74 | { |
| 75 | return rFm & 0x7fffffff; | 75 | return rFm & 0x7fffffff; |
| 76 | } | 76 | } |
| 77 | 77 | ||
| 78 | static float32 (*const monadic_single[16])(float32 rFm) = { | 78 | static float32 (*const monadic_single[16])(struct roundingData*, float32 rFm) = { |
| 79 | [MVF_CODE >> 20] = float32_mvf, | 79 | [MVF_CODE >> 20] = float32_mvf, |
| 80 | [MNF_CODE >> 20] = float32_mnf, | 80 | [MNF_CODE >> 20] = float32_mnf, |
| 81 | [ABS_CODE >> 20] = float32_abs, | 81 | [ABS_CODE >> 20] = float32_abs, |
| @@ -85,7 +85,7 @@ static float32 (*const monadic_single[16])(float32 rFm) = { | |||
| 85 | [NRM_CODE >> 20] = float32_mvf, | 85 | [NRM_CODE >> 20] = float32_mvf, |
| 86 | }; | 86 | }; |
| 87 | 87 | ||
| 88 | unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd) | 88 | unsigned int SingleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd) |
| 89 | { | 89 | { |
| 90 | FPA11 *fpa11 = GET_FPA11(); | 90 | FPA11 *fpa11 = GET_FPA11(); |
| 91 | float32 rFm; | 91 | float32 rFm; |
| @@ -108,13 +108,13 @@ unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd) | |||
| 108 | if (fpa11->fType[Fn] == typeSingle && | 108 | if (fpa11->fType[Fn] == typeSingle && |
| 109 | dyadic_single[opc_mask_shift]) { | 109 | dyadic_single[opc_mask_shift]) { |
| 110 | rFn = fpa11->fpreg[Fn].fSingle; | 110 | rFn = fpa11->fpreg[Fn].fSingle; |
| 111 | rFd->fSingle = dyadic_single[opc_mask_shift](rFn, rFm); | 111 | rFd->fSingle = dyadic_single[opc_mask_shift](roundData, rFn, rFm); |
| 112 | } else { | 112 | } else { |
| 113 | return 0; | 113 | return 0; |
| 114 | } | 114 | } |
| 115 | } else { | 115 | } else { |
| 116 | if (monadic_single[opc_mask_shift]) { | 116 | if (monadic_single[opc_mask_shift]) { |
| 117 | rFd->fSingle = monadic_single[opc_mask_shift](rFm); | 117 | rFd->fSingle = monadic_single[opc_mask_shift](roundData, rFm); |
| 118 | } else { | 118 | } else { |
| 119 | return 0; | 119 | return 0; |
| 120 | } | 120 | } |
diff --git a/arch/arm/nwfpe/softfloat.c b/arch/arm/nwfpe/softfloat.c index e038dd3be9b3..8b75a6e7cb3a 100644 --- a/arch/arm/nwfpe/softfloat.c +++ b/arch/arm/nwfpe/softfloat.c | |||
| @@ -36,16 +36,6 @@ this code that are retained. | |||
| 36 | 36 | ||
| 37 | /* | 37 | /* |
| 38 | ------------------------------------------------------------------------------- | 38 | ------------------------------------------------------------------------------- |
| 39 | Floating-point rounding mode, extended double-precision rounding precision, | ||
| 40 | and exception flags. | ||
| 41 | ------------------------------------------------------------------------------- | ||
| 42 | */ | ||
| 43 | int8 float_rounding_mode = float_round_nearest_even; | ||
| 44 | int8 floatx80_rounding_precision = 80; | ||
| 45 | int8 float_exception_flags; | ||
| 46 | |||
| 47 | /* | ||
| 48 | ------------------------------------------------------------------------------- | ||
| 49 | Primitive arithmetic functions, including multi-word arithmetic, and | 39 | Primitive arithmetic functions, including multi-word arithmetic, and |
| 50 | division and square root approximations. (Can be specialized to target if | 40 | division and square root approximations. (Can be specialized to target if |
| 51 | desired.) | 41 | desired.) |
| @@ -77,14 +67,14 @@ input is too large, however, the invalid exception is raised and the largest | |||
| 77 | positive or negative integer is returned. | 67 | positive or negative integer is returned. |
| 78 | ------------------------------------------------------------------------------- | 68 | ------------------------------------------------------------------------------- |
| 79 | */ | 69 | */ |
| 80 | static int32 roundAndPackInt32( flag zSign, bits64 absZ ) | 70 | static int32 roundAndPackInt32( struct roundingData *roundData, flag zSign, bits64 absZ ) |
| 81 | { | 71 | { |
| 82 | int8 roundingMode; | 72 | int8 roundingMode; |
| 83 | flag roundNearestEven; | 73 | flag roundNearestEven; |
| 84 | int8 roundIncrement, roundBits; | 74 | int8 roundIncrement, roundBits; |
| 85 | int32 z; | 75 | int32 z; |
| 86 | 76 | ||
| 87 | roundingMode = float_rounding_mode; | 77 | roundingMode = roundData->mode; |
| 88 | roundNearestEven = ( roundingMode == float_round_nearest_even ); | 78 | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
| 89 | roundIncrement = 0x40; | 79 | roundIncrement = 0x40; |
| 90 | if ( ! roundNearestEven ) { | 80 | if ( ! roundNearestEven ) { |
| @@ -107,10 +97,10 @@ static int32 roundAndPackInt32( flag zSign, bits64 absZ ) | |||
| 107 | z = absZ; | 97 | z = absZ; |
| 108 | if ( zSign ) z = - z; | 98 | if ( zSign ) z = - z; |
| 109 | if ( ( absZ>>32 ) || ( z && ( ( z < 0 ) ^ zSign ) ) ) { | 99 | if ( ( absZ>>32 ) || ( z && ( ( z < 0 ) ^ zSign ) ) ) { |
| 110 | float_exception_flags |= float_flag_invalid; | 100 | roundData->exception |= float_flag_invalid; |
| 111 | return zSign ? 0x80000000 : 0x7FFFFFFF; | 101 | return zSign ? 0x80000000 : 0x7FFFFFFF; |
| 112 | } | 102 | } |
| 113 | if ( roundBits ) float_exception_flags |= float_flag_inexact; | 103 | if ( roundBits ) roundData->exception |= float_flag_inexact; |
| 114 | return z; | 104 | return z; |
| 115 | 105 | ||
| 116 | } | 106 | } |
| @@ -224,14 +214,14 @@ The handling of underflow and overflow follows the IEC/IEEE Standard for | |||
| 224 | Binary Floating-point Arithmetic. | 214 | Binary Floating-point Arithmetic. |
| 225 | ------------------------------------------------------------------------------- | 215 | ------------------------------------------------------------------------------- |
| 226 | */ | 216 | */ |
| 227 | static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) | 217 | static float32 roundAndPackFloat32( struct roundingData *roundData, flag zSign, int16 zExp, bits32 zSig ) |
| 228 | { | 218 | { |
| 229 | int8 roundingMode; | 219 | int8 roundingMode; |
| 230 | flag roundNearestEven; | 220 | flag roundNearestEven; |
| 231 | int8 roundIncrement, roundBits; | 221 | int8 roundIncrement, roundBits; |
| 232 | flag isTiny; | 222 | flag isTiny; |
| 233 | 223 | ||
| 234 | roundingMode = float_rounding_mode; | 224 | roundingMode = roundData->mode; |
| 235 | roundNearestEven = ( roundingMode == float_round_nearest_even ); | 225 | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
| 236 | roundIncrement = 0x40; | 226 | roundIncrement = 0x40; |
| 237 | if ( ! roundNearestEven ) { | 227 | if ( ! roundNearestEven ) { |
| @@ -254,7 +244,7 @@ static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) | |||
| 254 | || ( ( zExp == 0xFD ) | 244 | || ( ( zExp == 0xFD ) |
| 255 | && ( (sbits32) ( zSig + roundIncrement ) < 0 ) ) | 245 | && ( (sbits32) ( zSig + roundIncrement ) < 0 ) ) |
| 256 | ) { | 246 | ) { |
| 257 | float_raise( float_flag_overflow | float_flag_inexact ); | 247 | roundData->exception |= float_flag_overflow | float_flag_inexact; |
| 258 | return packFloat32( zSign, 0xFF, 0 ) - ( roundIncrement == 0 ); | 248 | return packFloat32( zSign, 0xFF, 0 ) - ( roundIncrement == 0 ); |
| 259 | } | 249 | } |
| 260 | if ( zExp < 0 ) { | 250 | if ( zExp < 0 ) { |
| @@ -265,10 +255,10 @@ static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) | |||
| 265 | shift32RightJamming( zSig, - zExp, &zSig ); | 255 | shift32RightJamming( zSig, - zExp, &zSig ); |
| 266 | zExp = 0; | 256 | zExp = 0; |
| 267 | roundBits = zSig & 0x7F; | 257 | roundBits = zSig & 0x7F; |
| 268 | if ( isTiny && roundBits ) float_raise( float_flag_underflow ); | 258 | if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow; |
| 269 | } | 259 | } |
| 270 | } | 260 | } |
| 271 | if ( roundBits ) float_exception_flags |= float_flag_inexact; | 261 | if ( roundBits ) roundData->exception |= float_flag_inexact; |
| 272 | zSig = ( zSig + roundIncrement )>>7; | 262 | zSig = ( zSig + roundIncrement )>>7; |
| 273 | zSig &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); | 263 | zSig &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); |
| 274 | if ( zSig == 0 ) zExp = 0; | 264 | if ( zSig == 0 ) zExp = 0; |
| @@ -287,12 +277,12 @@ point exponent. | |||
| 287 | ------------------------------------------------------------------------------- | 277 | ------------------------------------------------------------------------------- |
| 288 | */ | 278 | */ |
| 289 | static float32 | 279 | static float32 |
| 290 | normalizeRoundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) | 280 | normalizeRoundAndPackFloat32( struct roundingData *roundData, flag zSign, int16 zExp, bits32 zSig ) |
| 291 | { | 281 | { |
| 292 | int8 shiftCount; | 282 | int8 shiftCount; |
| 293 | 283 | ||
| 294 | shiftCount = countLeadingZeros32( zSig ) - 1; | 284 | shiftCount = countLeadingZeros32( zSig ) - 1; |
| 295 | return roundAndPackFloat32( zSign, zExp - shiftCount, zSig<<shiftCount ); | 285 | return roundAndPackFloat32( roundData, zSign, zExp - shiftCount, zSig<<shiftCount ); |
| 296 | 286 | ||
| 297 | } | 287 | } |
| 298 | 288 | ||
| @@ -395,14 +385,14 @@ The handling of underflow and overflow follows the IEC/IEEE Standard for | |||
| 395 | Binary Floating-point Arithmetic. | 385 | Binary Floating-point Arithmetic. |
| 396 | ------------------------------------------------------------------------------- | 386 | ------------------------------------------------------------------------------- |
| 397 | */ | 387 | */ |
| 398 | static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) | 388 | static float64 roundAndPackFloat64( struct roundingData *roundData, flag zSign, int16 zExp, bits64 zSig ) |
| 399 | { | 389 | { |
| 400 | int8 roundingMode; | 390 | int8 roundingMode; |
| 401 | flag roundNearestEven; | 391 | flag roundNearestEven; |
| 402 | int16 roundIncrement, roundBits; | 392 | int16 roundIncrement, roundBits; |
| 403 | flag isTiny; | 393 | flag isTiny; |
| 404 | 394 | ||
| 405 | roundingMode = float_rounding_mode; | 395 | roundingMode = roundData->mode; |
| 406 | roundNearestEven = ( roundingMode == float_round_nearest_even ); | 396 | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
| 407 | roundIncrement = 0x200; | 397 | roundIncrement = 0x200; |
| 408 | if ( ! roundNearestEven ) { | 398 | if ( ! roundNearestEven ) { |
| @@ -427,7 +417,7 @@ static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) | |||
| 427 | ) { | 417 | ) { |
| 428 | //register int lr = __builtin_return_address(0); | 418 | //register int lr = __builtin_return_address(0); |
| 429 | //printk("roundAndPackFloat64 called from 0x%08x\n",lr); | 419 | //printk("roundAndPackFloat64 called from 0x%08x\n",lr); |
| 430 | float_raise( float_flag_overflow | float_flag_inexact ); | 420 | roundData->exception |= float_flag_overflow | float_flag_inexact; |
| 431 | return packFloat64( zSign, 0x7FF, 0 ) - ( roundIncrement == 0 ); | 421 | return packFloat64( zSign, 0x7FF, 0 ) - ( roundIncrement == 0 ); |
| 432 | } | 422 | } |
| 433 | if ( zExp < 0 ) { | 423 | if ( zExp < 0 ) { |
| @@ -438,10 +428,10 @@ static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) | |||
| 438 | shift64RightJamming( zSig, - zExp, &zSig ); | 428 | shift64RightJamming( zSig, - zExp, &zSig ); |
| 439 | zExp = 0; | 429 | zExp = 0; |
| 440 | roundBits = zSig & 0x3FF; | 430 | roundBits = zSig & 0x3FF; |
| 441 | if ( isTiny && roundBits ) float_raise( float_flag_underflow ); | 431 | if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow; |
| 442 | } | 432 | } |
| 443 | } | 433 | } |
| 444 | if ( roundBits ) float_exception_flags |= float_flag_inexact; | 434 | if ( roundBits ) roundData->exception |= float_flag_inexact; |
| 445 | zSig = ( zSig + roundIncrement )>>10; | 435 | zSig = ( zSig + roundIncrement )>>10; |
| 446 | zSig &= ~ ( ( ( roundBits ^ 0x200 ) == 0 ) & roundNearestEven ); | 436 | zSig &= ~ ( ( ( roundBits ^ 0x200 ) == 0 ) & roundNearestEven ); |
| 447 | if ( zSig == 0 ) zExp = 0; | 437 | if ( zSig == 0 ) zExp = 0; |
| @@ -460,12 +450,12 @@ point exponent. | |||
| 460 | ------------------------------------------------------------------------------- | 450 | ------------------------------------------------------------------------------- |
| 461 | */ | 451 | */ |
| 462 | static float64 | 452 | static float64 |
| 463 | normalizeRoundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) | 453 | normalizeRoundAndPackFloat64( struct roundingData *roundData, flag zSign, int16 zExp, bits64 zSig ) |
| 464 | { | 454 | { |
| 465 | int8 shiftCount; | 455 | int8 shiftCount; |
| 466 | 456 | ||
| 467 | shiftCount = countLeadingZeros64( zSig ) - 1; | 457 | shiftCount = countLeadingZeros64( zSig ) - 1; |
| 468 | return roundAndPackFloat64( zSign, zExp - shiftCount, zSig<<shiftCount ); | 458 | return roundAndPackFloat64( roundData, zSign, zExp - shiftCount, zSig<<shiftCount ); |
| 469 | 459 | ||
| 470 | } | 460 | } |
| 471 | 461 | ||
| @@ -572,14 +562,15 @@ Floating-point Arithmetic. | |||
| 572 | */ | 562 | */ |
| 573 | static floatx80 | 563 | static floatx80 |
| 574 | roundAndPackFloatx80( | 564 | roundAndPackFloatx80( |
| 575 | int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 | 565 | struct roundingData *roundData, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 |
| 576 | ) | 566 | ) |
| 577 | { | 567 | { |
| 578 | int8 roundingMode; | 568 | int8 roundingMode, roundingPrecision; |
| 579 | flag roundNearestEven, increment, isTiny; | 569 | flag roundNearestEven, increment, isTiny; |
| 580 | int64 roundIncrement, roundMask, roundBits; | 570 | int64 roundIncrement, roundMask, roundBits; |
| 581 | 571 | ||
| 582 | roundingMode = float_rounding_mode; | 572 | roundingMode = roundData->mode; |
| 573 | roundingPrecision = roundData->precision; | ||
| 583 | roundNearestEven = ( roundingMode == float_round_nearest_even ); | 574 | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
| 584 | if ( roundingPrecision == 80 ) goto precision80; | 575 | if ( roundingPrecision == 80 ) goto precision80; |
| 585 | if ( roundingPrecision == 64 ) { | 576 | if ( roundingPrecision == 64 ) { |
| @@ -623,8 +614,8 @@ static floatx80 | |||
| 623 | shift64RightJamming( zSig0, 1 - zExp, &zSig0 ); | 614 | shift64RightJamming( zSig0, 1 - zExp, &zSig0 ); |
| 624 | zExp = 0; | 615 | zExp = 0; |
| 625 | roundBits = zSig0 & roundMask; | 616 | roundBits = zSig0 & roundMask; |
| 626 | if ( isTiny && roundBits ) float_raise( float_flag_underflow ); | 617 | if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow; |
| 627 | if ( roundBits ) float_exception_flags |= float_flag_inexact; | 618 | if ( roundBits ) roundData->exception |= float_flag_inexact; |
| 628 | zSig0 += roundIncrement; | 619 | zSig0 += roundIncrement; |
| 629 | if ( (sbits64) zSig0 < 0 ) zExp = 1; | 620 | if ( (sbits64) zSig0 < 0 ) zExp = 1; |
| 630 | roundIncrement = roundMask + 1; | 621 | roundIncrement = roundMask + 1; |
| @@ -635,7 +626,7 @@ static floatx80 | |||
| 635 | return packFloatx80( zSign, zExp, zSig0 ); | 626 | return packFloatx80( zSign, zExp, zSig0 ); |
| 636 | } | 627 | } |
| 637 | } | 628 | } |
| 638 | if ( roundBits ) float_exception_flags |= float_flag_inexact; | 629 | if ( roundBits ) roundData->exception |= float_flag_inexact; |
| 639 | zSig0 += roundIncrement; | 630 | zSig0 += roundIncrement; |
| 640 | if ( zSig0 < roundIncrement ) { | 631 | if ( zSig0 < roundIncrement ) { |
| 641 | ++zExp; | 632 | ++zExp; |
| @@ -672,7 +663,7 @@ static floatx80 | |||
| 672 | ) { | 663 | ) { |
| 673 | roundMask = 0; | 664 | roundMask = 0; |
| 674 | overflow: | 665 | overflow: |
| 675 | float_raise( float_flag_overflow | float_flag_inexact ); | 666 | roundData->exception |= float_flag_overflow | float_flag_inexact; |
| 676 | if ( ( roundingMode == float_round_to_zero ) | 667 | if ( ( roundingMode == float_round_to_zero ) |
| 677 | || ( zSign && ( roundingMode == float_round_up ) ) | 668 | || ( zSign && ( roundingMode == float_round_up ) ) |
| 678 | || ( ! zSign && ( roundingMode == float_round_down ) ) | 669 | || ( ! zSign && ( roundingMode == float_round_down ) ) |
| @@ -689,8 +680,8 @@ static floatx80 | |||
| 689 | || ( zSig0 < LIT64( 0xFFFFFFFFFFFFFFFF ) ); | 680 | || ( zSig0 < LIT64( 0xFFFFFFFFFFFFFFFF ) ); |
| 690 | shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 ); | 681 | shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 ); |
| 691 | zExp = 0; | 682 | zExp = 0; |
| 692 | if ( isTiny && zSig1 ) float_raise( float_flag_underflow ); | 683 | if ( isTiny && zSig1 ) roundData->exception |= float_flag_underflow; |
| 693 | if ( zSig1 ) float_exception_flags |= float_flag_inexact; | 684 | if ( zSig1 ) roundData->exception |= float_flag_inexact; |
| 694 | if ( roundNearestEven ) { | 685 | if ( roundNearestEven ) { |
| 695 | increment = ( (sbits64) zSig1 < 0 ); | 686 | increment = ( (sbits64) zSig1 < 0 ); |
| 696 | } | 687 | } |
| @@ -710,7 +701,7 @@ static floatx80 | |||
| 710 | return packFloatx80( zSign, zExp, zSig0 ); | 701 | return packFloatx80( zSign, zExp, zSig0 ); |
| 711 | } | 702 | } |
| 712 | } | 703 | } |
| 713 | if ( zSig1 ) float_exception_flags |= float_flag_inexact; | 704 | if ( zSig1 ) roundData->exception |= float_flag_inexact; |
| 714 | if ( increment ) { | 705 | if ( increment ) { |
| 715 | ++zSig0; | 706 | ++zSig0; |
| 716 | if ( zSig0 == 0 ) { | 707 | if ( zSig0 == 0 ) { |
| @@ -740,7 +731,7 @@ normalized. | |||
| 740 | */ | 731 | */ |
| 741 | static floatx80 | 732 | static floatx80 |
| 742 | normalizeRoundAndPackFloatx80( | 733 | normalizeRoundAndPackFloatx80( |
| 743 | int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 | 734 | struct roundingData *roundData, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 |
| 744 | ) | 735 | ) |
| 745 | { | 736 | { |
| 746 | int8 shiftCount; | 737 | int8 shiftCount; |
| @@ -754,7 +745,7 @@ static floatx80 | |||
| 754 | shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); | 745 | shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); |
| 755 | zExp -= shiftCount; | 746 | zExp -= shiftCount; |
| 756 | return | 747 | return |
| 757 | roundAndPackFloatx80( roundingPrecision, zSign, zExp, zSig0, zSig1 ); | 748 | roundAndPackFloatx80( roundData, zSign, zExp, zSig0, zSig1 ); |
| 758 | 749 | ||
| 759 | } | 750 | } |
| 760 | 751 | ||
| @@ -767,14 +758,14 @@ the single-precision floating-point format. The conversion is performed | |||
| 767 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 758 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 768 | ------------------------------------------------------------------------------- | 759 | ------------------------------------------------------------------------------- |
| 769 | */ | 760 | */ |
| 770 | float32 int32_to_float32( int32 a ) | 761 | float32 int32_to_float32(struct roundingData *roundData, int32 a) |
| 771 | { | 762 | { |
| 772 | flag zSign; | 763 | flag zSign; |
| 773 | 764 | ||
| 774 | if ( a == 0 ) return 0; | 765 | if ( a == 0 ) return 0; |
| 775 | if ( a == 0x80000000 ) return packFloat32( 1, 0x9E, 0 ); | 766 | if ( a == 0x80000000 ) return packFloat32( 1, 0x9E, 0 ); |
| 776 | zSign = ( a < 0 ); | 767 | zSign = ( a < 0 ); |
| 777 | return normalizeRoundAndPackFloat32( zSign, 0x9C, zSign ? - a : a ); | 768 | return normalizeRoundAndPackFloat32( roundData, zSign, 0x9C, zSign ? - a : a ); |
| 778 | 769 | ||
| 779 | } | 770 | } |
| 780 | 771 | ||
| @@ -840,7 +831,7 @@ positive integer is returned. Otherwise, if the conversion overflows, the | |||
| 840 | largest integer with the same sign as `a' is returned. | 831 | largest integer with the same sign as `a' is returned. |
| 841 | ------------------------------------------------------------------------------- | 832 | ------------------------------------------------------------------------------- |
| 842 | */ | 833 | */ |
| 843 | int32 float32_to_int32( float32 a ) | 834 | int32 float32_to_int32( struct roundingData *roundData, float32 a ) |
| 844 | { | 835 | { |
| 845 | flag aSign; | 836 | flag aSign; |
| 846 | int16 aExp, shiftCount; | 837 | int16 aExp, shiftCount; |
| @@ -856,7 +847,7 @@ int32 float32_to_int32( float32 a ) | |||
| 856 | zSig = aSig; | 847 | zSig = aSig; |
| 857 | zSig <<= 32; | 848 | zSig <<= 32; |
| 858 | if ( 0 < shiftCount ) shift64RightJamming( zSig, shiftCount, &zSig ); | 849 | if ( 0 < shiftCount ) shift64RightJamming( zSig, shiftCount, &zSig ); |
| 859 | return roundAndPackInt32( aSign, zSig ); | 850 | return roundAndPackInt32( roundData, aSign, zSig ); |
| 860 | 851 | ||
| 861 | } | 852 | } |
| 862 | 853 | ||
| @@ -889,13 +880,13 @@ int32 float32_to_int32_round_to_zero( float32 a ) | |||
| 889 | return 0x80000000; | 880 | return 0x80000000; |
| 890 | } | 881 | } |
| 891 | else if ( aExp <= 0x7E ) { | 882 | else if ( aExp <= 0x7E ) { |
| 892 | if ( aExp | aSig ) float_exception_flags |= float_flag_inexact; | 883 | if ( aExp | aSig ) float_raise( float_flag_inexact ); |
| 893 | return 0; | 884 | return 0; |
| 894 | } | 885 | } |
| 895 | aSig = ( aSig | 0x00800000 )<<8; | 886 | aSig = ( aSig | 0x00800000 )<<8; |
| 896 | z = aSig>>( - shiftCount ); | 887 | z = aSig>>( - shiftCount ); |
| 897 | if ( (bits32) ( aSig<<( shiftCount & 31 ) ) ) { | 888 | if ( (bits32) ( aSig<<( shiftCount & 31 ) ) ) { |
| 898 | float_exception_flags |= float_flag_inexact; | 889 | float_raise( float_flag_inexact ); |
| 899 | } | 890 | } |
| 900 | return aSign ? - z : z; | 891 | return aSign ? - z : z; |
| 901 | 892 | ||
| @@ -973,7 +964,7 @@ operation is performed according to the IEC/IEEE Standard for Binary | |||
| 973 | Floating-point Arithmetic. | 964 | Floating-point Arithmetic. |
| 974 | ------------------------------------------------------------------------------- | 965 | ------------------------------------------------------------------------------- |
| 975 | */ | 966 | */ |
| 976 | float32 float32_round_to_int( float32 a ) | 967 | float32 float32_round_to_int( struct roundingData *roundData, float32 a ) |
| 977 | { | 968 | { |
| 978 | flag aSign; | 969 | flag aSign; |
| 979 | int16 aExp; | 970 | int16 aExp; |
| @@ -988,11 +979,12 @@ float32 float32_round_to_int( float32 a ) | |||
| 988 | } | 979 | } |
| 989 | return a; | 980 | return a; |
| 990 | } | 981 | } |
| 982 | roundingMode = roundData->mode; | ||
| 991 | if ( aExp <= 0x7E ) { | 983 | if ( aExp <= 0x7E ) { |
| 992 | if ( (bits32) ( a<<1 ) == 0 ) return a; | 984 | if ( (bits32) ( a<<1 ) == 0 ) return a; |
| 993 | float_exception_flags |= float_flag_inexact; | 985 | roundData->exception |= float_flag_inexact; |
| 994 | aSign = extractFloat32Sign( a ); | 986 | aSign = extractFloat32Sign( a ); |
| 995 | switch ( float_rounding_mode ) { | 987 | switch ( roundingMode ) { |
| 996 | case float_round_nearest_even: | 988 | case float_round_nearest_even: |
| 997 | if ( ( aExp == 0x7E ) && extractFloat32Frac( a ) ) { | 989 | if ( ( aExp == 0x7E ) && extractFloat32Frac( a ) ) { |
| 998 | return packFloat32( aSign, 0x7F, 0 ); | 990 | return packFloat32( aSign, 0x7F, 0 ); |
| @@ -1009,7 +1001,6 @@ float32 float32_round_to_int( float32 a ) | |||
| 1009 | lastBitMask <<= 0x96 - aExp; | 1001 | lastBitMask <<= 0x96 - aExp; |
| 1010 | roundBitsMask = lastBitMask - 1; | 1002 | roundBitsMask = lastBitMask - 1; |
| 1011 | z = a; | 1003 | z = a; |
| 1012 | roundingMode = float_rounding_mode; | ||
| 1013 | if ( roundingMode == float_round_nearest_even ) { | 1004 | if ( roundingMode == float_round_nearest_even ) { |
| 1014 | z += lastBitMask>>1; | 1005 | z += lastBitMask>>1; |
| 1015 | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; | 1006 | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; |
| @@ -1020,7 +1011,7 @@ float32 float32_round_to_int( float32 a ) | |||
| 1020 | } | 1011 | } |
| 1021 | } | 1012 | } |
| 1022 | z &= ~ roundBitsMask; | 1013 | z &= ~ roundBitsMask; |
| 1023 | if ( z != a ) float_exception_flags |= float_flag_inexact; | 1014 | if ( z != a ) roundData->exception |= float_flag_inexact; |
| 1024 | return z; | 1015 | return z; |
| 1025 | 1016 | ||
| 1026 | } | 1017 | } |
| @@ -1034,7 +1025,7 @@ addition is performed according to the IEC/IEEE Standard for Binary | |||
| 1034 | Floating-point Arithmetic. | 1025 | Floating-point Arithmetic. |
| 1035 | ------------------------------------------------------------------------------- | 1026 | ------------------------------------------------------------------------------- |
| 1036 | */ | 1027 | */ |
| 1037 | static float32 addFloat32Sigs( float32 a, float32 b, flag zSign ) | 1028 | static float32 addFloat32Sigs( struct roundingData *roundData, float32 a, float32 b, flag zSign ) |
| 1038 | { | 1029 | { |
| 1039 | int16 aExp, bExp, zExp; | 1030 | int16 aExp, bExp, zExp; |
| 1040 | bits32 aSig, bSig, zSig; | 1031 | bits32 aSig, bSig, zSig; |
| @@ -1093,7 +1084,7 @@ static float32 addFloat32Sigs( float32 a, float32 b, flag zSign ) | |||
| 1093 | ++zExp; | 1084 | ++zExp; |
| 1094 | } | 1085 | } |
| 1095 | roundAndPack: | 1086 | roundAndPack: |
| 1096 | return roundAndPackFloat32( zSign, zExp, zSig ); | 1087 | return roundAndPackFloat32( roundData, zSign, zExp, zSig ); |
| 1097 | 1088 | ||
| 1098 | } | 1089 | } |
| 1099 | 1090 | ||
| @@ -1106,7 +1097,7 @@ result is a NaN. The subtraction is performed according to the IEC/IEEE | |||
| 1106 | Standard for Binary Floating-point Arithmetic. | 1097 | Standard for Binary Floating-point Arithmetic. |
| 1107 | ------------------------------------------------------------------------------- | 1098 | ------------------------------------------------------------------------------- |
| 1108 | */ | 1099 | */ |
| 1109 | static float32 subFloat32Sigs( float32 a, float32 b, flag zSign ) | 1100 | static float32 subFloat32Sigs( struct roundingData *roundData, float32 a, float32 b, flag zSign ) |
| 1110 | { | 1101 | { |
| 1111 | int16 aExp, bExp, zExp; | 1102 | int16 aExp, bExp, zExp; |
| 1112 | bits32 aSig, bSig, zSig; | 1103 | bits32 aSig, bSig, zSig; |
| @@ -1123,7 +1114,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign ) | |||
| 1123 | if ( expDiff < 0 ) goto bExpBigger; | 1114 | if ( expDiff < 0 ) goto bExpBigger; |
| 1124 | if ( aExp == 0xFF ) { | 1115 | if ( aExp == 0xFF ) { |
| 1125 | if ( aSig | bSig ) return propagateFloat32NaN( a, b ); | 1116 | if ( aSig | bSig ) return propagateFloat32NaN( a, b ); |
| 1126 | float_raise( float_flag_invalid ); | 1117 | roundData->exception |= float_flag_invalid; |
| 1127 | return float32_default_nan; | 1118 | return float32_default_nan; |
| 1128 | } | 1119 | } |
| 1129 | if ( aExp == 0 ) { | 1120 | if ( aExp == 0 ) { |
| @@ -1132,7 +1123,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign ) | |||
| 1132 | } | 1123 | } |
| 1133 | if ( bSig < aSig ) goto aBigger; | 1124 | if ( bSig < aSig ) goto aBigger; |
| 1134 | if ( aSig < bSig ) goto bBigger; | 1125 | if ( aSig < bSig ) goto bBigger; |
| 1135 | return packFloat32( float_rounding_mode == float_round_down, 0, 0 ); | 1126 | return packFloat32( roundData->mode == float_round_down, 0, 0 ); |
| 1136 | bExpBigger: | 1127 | bExpBigger: |
| 1137 | if ( bExp == 0xFF ) { | 1128 | if ( bExp == 0xFF ) { |
| 1138 | if ( bSig ) return propagateFloat32NaN( a, b ); | 1129 | if ( bSig ) return propagateFloat32NaN( a, b ); |
| @@ -1169,7 +1160,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign ) | |||
| 1169 | zExp = aExp; | 1160 | zExp = aExp; |
| 1170 | normalizeRoundAndPack: | 1161 | normalizeRoundAndPack: |
| 1171 | --zExp; | 1162 | --zExp; |
| 1172 | return normalizeRoundAndPackFloat32( zSign, zExp, zSig ); | 1163 | return normalizeRoundAndPackFloat32( roundData, zSign, zExp, zSig ); |
| 1173 | 1164 | ||
| 1174 | } | 1165 | } |
| 1175 | 1166 | ||
| @@ -1180,17 +1171,17 @@ and `b'. The operation is performed according to the IEC/IEEE Standard for | |||
| 1180 | Binary Floating-point Arithmetic. | 1171 | Binary Floating-point Arithmetic. |
| 1181 | ------------------------------------------------------------------------------- | 1172 | ------------------------------------------------------------------------------- |
| 1182 | */ | 1173 | */ |
| 1183 | float32 float32_add( float32 a, float32 b ) | 1174 | float32 float32_add( struct roundingData *roundData, float32 a, float32 b ) |
| 1184 | { | 1175 | { |
| 1185 | flag aSign, bSign; | 1176 | flag aSign, bSign; |
| 1186 | 1177 | ||
| 1187 | aSign = extractFloat32Sign( a ); | 1178 | aSign = extractFloat32Sign( a ); |
| 1188 | bSign = extractFloat32Sign( b ); | 1179 | bSign = extractFloat32Sign( b ); |
| 1189 | if ( aSign == bSign ) { | 1180 | if ( aSign == bSign ) { |
| 1190 | return addFloat32Sigs( a, b, aSign ); | 1181 | return addFloat32Sigs( roundData, a, b, aSign ); |
| 1191 | } | 1182 | } |
| 1192 | else { | 1183 | else { |
| 1193 | return subFloat32Sigs( a, b, aSign ); | 1184 | return subFloat32Sigs( roundData, a, b, aSign ); |
| 1194 | } | 1185 | } |
| 1195 | 1186 | ||
| 1196 | } | 1187 | } |
| @@ -1202,17 +1193,17 @@ Returns the result of subtracting the single-precision floating-point values | |||
| 1202 | for Binary Floating-point Arithmetic. | 1193 | for Binary Floating-point Arithmetic. |
| 1203 | ------------------------------------------------------------------------------- | 1194 | ------------------------------------------------------------------------------- |
| 1204 | */ | 1195 | */ |
| 1205 | float32 float32_sub( float32 a, float32 b ) | 1196 | float32 float32_sub( struct roundingData *roundData, float32 a, float32 b ) |
| 1206 | { | 1197 | { |
| 1207 | flag aSign, bSign; | 1198 | flag aSign, bSign; |
| 1208 | 1199 | ||
| 1209 | aSign = extractFloat32Sign( a ); | 1200 | aSign = extractFloat32Sign( a ); |
| 1210 | bSign = extractFloat32Sign( b ); | 1201 | bSign = extractFloat32Sign( b ); |
| 1211 | if ( aSign == bSign ) { | 1202 | if ( aSign == bSign ) { |
| 1212 | return subFloat32Sigs( a, b, aSign ); | 1203 | return subFloat32Sigs( roundData, a, b, aSign ); |
| 1213 | } | 1204 | } |
| 1214 | else { | 1205 | else { |
| 1215 | return addFloat32Sigs( a, b, aSign ); | 1206 | return addFloat32Sigs( roundData, a, b, aSign ); |
| 1216 | } | 1207 | } |
| 1217 | 1208 | ||
| 1218 | } | 1209 | } |
| @@ -1224,7 +1215,7 @@ Returns the result of multiplying the single-precision floating-point values | |||
| 1224 | for Binary Floating-point Arithmetic. | 1215 | for Binary Floating-point Arithmetic. |
| 1225 | ------------------------------------------------------------------------------- | 1216 | ------------------------------------------------------------------------------- |
| 1226 | */ | 1217 | */ |
| 1227 | float32 float32_mul( float32 a, float32 b ) | 1218 | float32 float32_mul( struct roundingData *roundData, float32 a, float32 b ) |
| 1228 | { | 1219 | { |
| 1229 | flag aSign, bSign, zSign; | 1220 | flag aSign, bSign, zSign; |
| 1230 | int16 aExp, bExp, zExp; | 1221 | int16 aExp, bExp, zExp; |
| @@ -1244,7 +1235,7 @@ float32 float32_mul( float32 a, float32 b ) | |||
| 1244 | return propagateFloat32NaN( a, b ); | 1235 | return propagateFloat32NaN( a, b ); |
| 1245 | } | 1236 | } |
| 1246 | if ( ( bExp | bSig ) == 0 ) { | 1237 | if ( ( bExp | bSig ) == 0 ) { |
| 1247 | float_raise( float_flag_invalid ); | 1238 | roundData->exception |= float_flag_invalid; |
| 1248 | return float32_default_nan; | 1239 | return float32_default_nan; |
| 1249 | } | 1240 | } |
| 1250 | return packFloat32( zSign, 0xFF, 0 ); | 1241 | return packFloat32( zSign, 0xFF, 0 ); |
| @@ -1252,7 +1243,7 @@ float32 float32_mul( float32 a, float32 b ) | |||
| 1252 | if ( bExp == 0xFF ) { | 1243 | if ( bExp == 0xFF ) { |
| 1253 | if ( bSig ) return propagateFloat32NaN( a, b ); | 1244 | if ( bSig ) return propagateFloat32NaN( a, b ); |
| 1254 | if ( ( aExp | aSig ) == 0 ) { | 1245 | if ( ( aExp | aSig ) == 0 ) { |
| 1255 | float_raise( float_flag_invalid ); | 1246 | roundData->exception |= float_flag_invalid; |
| 1256 | return float32_default_nan; | 1247 | return float32_default_nan; |
| 1257 | } | 1248 | } |
| 1258 | return packFloat32( zSign, 0xFF, 0 ); | 1249 | return packFloat32( zSign, 0xFF, 0 ); |
| @@ -1274,7 +1265,7 @@ float32 float32_mul( float32 a, float32 b ) | |||
| 1274 | zSig <<= 1; | 1265 | zSig <<= 1; |
| 1275 | --zExp; | 1266 | --zExp; |
| 1276 | } | 1267 | } |
| 1277 | return roundAndPackFloat32( zSign, zExp, zSig ); | 1268 | return roundAndPackFloat32( roundData, zSign, zExp, zSig ); |
| 1278 | 1269 | ||
| 1279 | } | 1270 | } |
| 1280 | 1271 | ||
| @@ -1285,7 +1276,7 @@ by the corresponding value `b'. The operation is performed according to the | |||
| 1285 | IEC/IEEE Standard for Binary Floating-point Arithmetic. | 1276 | IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 1286 | ------------------------------------------------------------------------------- | 1277 | ------------------------------------------------------------------------------- |
| 1287 | */ | 1278 | */ |
| 1288 | float32 float32_div( float32 a, float32 b ) | 1279 | float32 float32_div( struct roundingData *roundData, float32 a, float32 b ) |
| 1289 | { | 1280 | { |
| 1290 | flag aSign, bSign, zSign; | 1281 | flag aSign, bSign, zSign; |
| 1291 | int16 aExp, bExp, zExp; | 1282 | int16 aExp, bExp, zExp; |
| @@ -1302,7 +1293,7 @@ float32 float32_div( float32 a, float32 b ) | |||
| 1302 | if ( aSig ) return propagateFloat32NaN( a, b ); | 1293 | if ( aSig ) return propagateFloat32NaN( a, b ); |
| 1303 | if ( bExp == 0xFF ) { | 1294 | if ( bExp == 0xFF ) { |
| 1304 | if ( bSig ) return propagateFloat32NaN( a, b ); | 1295 | if ( bSig ) return propagateFloat32NaN( a, b ); |
| 1305 | float_raise( float_flag_invalid ); | 1296 | roundData->exception |= float_flag_invalid; |
| 1306 | return float32_default_nan; | 1297 | return float32_default_nan; |
| 1307 | } | 1298 | } |
| 1308 | return packFloat32( zSign, 0xFF, 0 ); | 1299 | return packFloat32( zSign, 0xFF, 0 ); |
| @@ -1314,10 +1305,10 @@ float32 float32_div( float32 a, float32 b ) | |||
| 1314 | if ( bExp == 0 ) { | 1305 | if ( bExp == 0 ) { |
| 1315 | if ( bSig == 0 ) { | 1306 | if ( bSig == 0 ) { |
| 1316 | if ( ( aExp | aSig ) == 0 ) { | 1307 | if ( ( aExp | aSig ) == 0 ) { |
| 1317 | float_raise( float_flag_invalid ); | 1308 | roundData->exception |= float_flag_invalid; |
| 1318 | return float32_default_nan; | 1309 | return float32_default_nan; |
| 1319 | } | 1310 | } |
| 1320 | float_raise( float_flag_divbyzero ); | 1311 | roundData->exception |= float_flag_divbyzero; |
| 1321 | return packFloat32( zSign, 0xFF, 0 ); | 1312 | return packFloat32( zSign, 0xFF, 0 ); |
| 1322 | } | 1313 | } |
| 1323 | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); | 1314 | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); |
| @@ -1341,7 +1332,7 @@ float32 float32_div( float32 a, float32 b ) | |||
| 1341 | if ( ( zSig & 0x3F ) == 0 ) { | 1332 | if ( ( zSig & 0x3F ) == 0 ) { |
| 1342 | zSig |= ( ( (bits64) bSig ) * zSig != ( (bits64) aSig )<<32 ); | 1333 | zSig |= ( ( (bits64) bSig ) * zSig != ( (bits64) aSig )<<32 ); |
| 1343 | } | 1334 | } |
| 1344 | return roundAndPackFloat32( zSign, zExp, zSig ); | 1335 | return roundAndPackFloat32( roundData, zSign, zExp, zSig ); |
| 1345 | 1336 | ||
| 1346 | } | 1337 | } |
| 1347 | 1338 | ||
| @@ -1352,7 +1343,7 @@ with respect to the corresponding value `b'. The operation is performed | |||
| 1352 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 1343 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 1353 | ------------------------------------------------------------------------------- | 1344 | ------------------------------------------------------------------------------- |
| 1354 | */ | 1345 | */ |
| 1355 | float32 float32_rem( float32 a, float32 b ) | 1346 | float32 float32_rem( struct roundingData *roundData, float32 a, float32 b ) |
| 1356 | { | 1347 | { |
| 1357 | flag aSign, bSign, zSign; | 1348 | flag aSign, bSign, zSign; |
| 1358 | int16 aExp, bExp, expDiff; | 1349 | int16 aExp, bExp, expDiff; |
| @@ -1372,7 +1363,7 @@ float32 float32_rem( float32 a, float32 b ) | |||
| 1372 | if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { | 1363 | if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { |
| 1373 | return propagateFloat32NaN( a, b ); | 1364 | return propagateFloat32NaN( a, b ); |
| 1374 | } | 1365 | } |
| 1375 | float_raise( float_flag_invalid ); | 1366 | roundData->exception |= float_flag_invalid; |
| 1376 | return float32_default_nan; | 1367 | return float32_default_nan; |
| 1377 | } | 1368 | } |
| 1378 | if ( bExp == 0xFF ) { | 1369 | if ( bExp == 0xFF ) { |
| @@ -1381,7 +1372,7 @@ float32 float32_rem( float32 a, float32 b ) | |||
| 1381 | } | 1372 | } |
| 1382 | if ( bExp == 0 ) { | 1373 | if ( bExp == 0 ) { |
| 1383 | if ( bSig == 0 ) { | 1374 | if ( bSig == 0 ) { |
| 1384 | float_raise( float_flag_invalid ); | 1375 | roundData->exception |= float_flag_invalid; |
| 1385 | return float32_default_nan; | 1376 | return float32_default_nan; |
| 1386 | } | 1377 | } |
| 1387 | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); | 1378 | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); |
| @@ -1444,7 +1435,7 @@ float32 float32_rem( float32 a, float32 b ) | |||
| 1444 | } | 1435 | } |
| 1445 | zSign = ( (sbits32) aSig < 0 ); | 1436 | zSign = ( (sbits32) aSig < 0 ); |
| 1446 | if ( zSign ) aSig = - aSig; | 1437 | if ( zSign ) aSig = - aSig; |
| 1447 | return normalizeRoundAndPackFloat32( aSign ^ zSign, bExp, aSig ); | 1438 | return normalizeRoundAndPackFloat32( roundData, aSign ^ zSign, bExp, aSig ); |
| 1448 | 1439 | ||
| 1449 | } | 1440 | } |
| 1450 | 1441 | ||
| @@ -1455,7 +1446,7 @@ The operation is performed according to the IEC/IEEE Standard for Binary | |||
| 1455 | Floating-point Arithmetic. | 1446 | Floating-point Arithmetic. |
| 1456 | ------------------------------------------------------------------------------- | 1447 | ------------------------------------------------------------------------------- |
| 1457 | */ | 1448 | */ |
| 1458 | float32 float32_sqrt( float32 a ) | 1449 | float32 float32_sqrt( struct roundingData *roundData, float32 a ) |
| 1459 | { | 1450 | { |
| 1460 | flag aSign; | 1451 | flag aSign; |
| 1461 | int16 aExp, zExp; | 1452 | int16 aExp, zExp; |
| @@ -1468,12 +1459,12 @@ float32 float32_sqrt( float32 a ) | |||
| 1468 | if ( aExp == 0xFF ) { | 1459 | if ( aExp == 0xFF ) { |
| 1469 | if ( aSig ) return propagateFloat32NaN( a, 0 ); | 1460 | if ( aSig ) return propagateFloat32NaN( a, 0 ); |
| 1470 | if ( ! aSign ) return a; | 1461 | if ( ! aSign ) return a; |
| 1471 | float_raise( float_flag_invalid ); | 1462 | roundData->exception |= float_flag_invalid; |
| 1472 | return float32_default_nan; | 1463 | return float32_default_nan; |
| 1473 | } | 1464 | } |
| 1474 | if ( aSign ) { | 1465 | if ( aSign ) { |
| 1475 | if ( ( aExp | aSig ) == 0 ) return a; | 1466 | if ( ( aExp | aSig ) == 0 ) return a; |
| 1476 | float_raise( float_flag_invalid ); | 1467 | roundData->exception |= float_flag_invalid; |
| 1477 | return float32_default_nan; | 1468 | return float32_default_nan; |
| 1478 | } | 1469 | } |
| 1479 | if ( aExp == 0 ) { | 1470 | if ( aExp == 0 ) { |
| @@ -1499,7 +1490,7 @@ float32 float32_sqrt( float32 a ) | |||
| 1499 | } | 1490 | } |
| 1500 | } | 1491 | } |
| 1501 | shift32RightJamming( zSig, 1, &zSig ); | 1492 | shift32RightJamming( zSig, 1, &zSig ); |
| 1502 | return roundAndPackFloat32( 0, zExp, zSig ); | 1493 | return roundAndPackFloat32( roundData, 0, zExp, zSig ); |
| 1503 | 1494 | ||
| 1504 | } | 1495 | } |
| 1505 | 1496 | ||
| @@ -1661,7 +1652,7 @@ positive integer is returned. Otherwise, if the conversion overflows, the | |||
| 1661 | largest integer with the same sign as `a' is returned. | 1652 | largest integer with the same sign as `a' is returned. |
| 1662 | ------------------------------------------------------------------------------- | 1653 | ------------------------------------------------------------------------------- |
| 1663 | */ | 1654 | */ |
| 1664 | int32 float64_to_int32( float64 a ) | 1655 | int32 float64_to_int32( struct roundingData *roundData, float64 a ) |
| 1665 | { | 1656 | { |
| 1666 | flag aSign; | 1657 | flag aSign; |
| 1667 | int16 aExp, shiftCount; | 1658 | int16 aExp, shiftCount; |
| @@ -1674,7 +1665,7 @@ int32 float64_to_int32( float64 a ) | |||
| 1674 | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); | 1665 | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); |
| 1675 | shiftCount = 0x42C - aExp; | 1666 | shiftCount = 0x42C - aExp; |
| 1676 | if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); | 1667 | if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); |
| 1677 | return roundAndPackInt32( aSign, aSig ); | 1668 | return roundAndPackInt32( roundData, aSign, aSig ); |
| 1678 | 1669 | ||
| 1679 | } | 1670 | } |
| 1680 | 1671 | ||
| @@ -1705,7 +1696,7 @@ int32 float64_to_int32_round_to_zero( float64 a ) | |||
| 1705 | goto invalid; | 1696 | goto invalid; |
| 1706 | } | 1697 | } |
| 1707 | else if ( 52 < shiftCount ) { | 1698 | else if ( 52 < shiftCount ) { |
| 1708 | if ( aExp || aSig ) float_exception_flags |= float_flag_inexact; | 1699 | if ( aExp || aSig ) float_raise( float_flag_inexact ); |
| 1709 | return 0; | 1700 | return 0; |
| 1710 | } | 1701 | } |
| 1711 | aSig |= LIT64( 0x0010000000000000 ); | 1702 | aSig |= LIT64( 0x0010000000000000 ); |
| @@ -1715,11 +1706,11 @@ int32 float64_to_int32_round_to_zero( float64 a ) | |||
| 1715 | if ( aSign ) z = - z; | 1706 | if ( aSign ) z = - z; |
| 1716 | if ( ( z < 0 ) ^ aSign ) { | 1707 | if ( ( z < 0 ) ^ aSign ) { |
| 1717 | invalid: | 1708 | invalid: |
| 1718 | float_exception_flags |= float_flag_invalid; | 1709 | float_raise( float_flag_invalid ); |
| 1719 | return aSign ? 0x80000000 : 0x7FFFFFFF; | 1710 | return aSign ? 0x80000000 : 0x7FFFFFFF; |
| 1720 | } | 1711 | } |
| 1721 | if ( ( aSig<<shiftCount ) != savedASig ) { | 1712 | if ( ( aSig<<shiftCount ) != savedASig ) { |
| 1722 | float_exception_flags |= float_flag_inexact; | 1713 | float_raise( float_flag_inexact ); |
| 1723 | } | 1714 | } |
| 1724 | return z; | 1715 | return z; |
| 1725 | 1716 | ||
| @@ -1736,7 +1727,7 @@ positive integer is returned. Otherwise, if the conversion overflows, the | |||
| 1736 | largest positive integer is returned. | 1727 | largest positive integer is returned. |
| 1737 | ------------------------------------------------------------------------------- | 1728 | ------------------------------------------------------------------------------- |
| 1738 | */ | 1729 | */ |
| 1739 | int32 float64_to_uint32( float64 a ) | 1730 | int32 float64_to_uint32( struct roundingData *roundData, float64 a ) |
| 1740 | { | 1731 | { |
| 1741 | flag aSign; | 1732 | flag aSign; |
| 1742 | int16 aExp, shiftCount; | 1733 | int16 aExp, shiftCount; |
| @@ -1749,7 +1740,7 @@ int32 float64_to_uint32( float64 a ) | |||
| 1749 | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); | 1740 | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); |
| 1750 | shiftCount = 0x42C - aExp; | 1741 | shiftCount = 0x42C - aExp; |
| 1751 | if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); | 1742 | if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); |
| 1752 | return roundAndPackInt32( aSign, aSig ); | 1743 | return roundAndPackInt32( roundData, aSign, aSig ); |
| 1753 | } | 1744 | } |
| 1754 | 1745 | ||
| 1755 | /* | 1746 | /* |
| @@ -1778,7 +1769,7 @@ int32 float64_to_uint32_round_to_zero( float64 a ) | |||
| 1778 | goto invalid; | 1769 | goto invalid; |
| 1779 | } | 1770 | } |
| 1780 | else if ( 52 < shiftCount ) { | 1771 | else if ( 52 < shiftCount ) { |
| 1781 | if ( aExp || aSig ) float_exception_flags |= float_flag_inexact; | 1772 | if ( aExp || aSig ) float_raise( float_flag_inexact ); |
| 1782 | return 0; | 1773 | return 0; |
| 1783 | } | 1774 | } |
| 1784 | aSig |= LIT64( 0x0010000000000000 ); | 1775 | aSig |= LIT64( 0x0010000000000000 ); |
| @@ -1788,11 +1779,11 @@ int32 float64_to_uint32_round_to_zero( float64 a ) | |||
| 1788 | if ( aSign ) z = - z; | 1779 | if ( aSign ) z = - z; |
| 1789 | if ( ( z < 0 ) ^ aSign ) { | 1780 | if ( ( z < 0 ) ^ aSign ) { |
| 1790 | invalid: | 1781 | invalid: |
| 1791 | float_exception_flags |= float_flag_invalid; | 1782 | float_raise( float_flag_invalid ); |
| 1792 | return aSign ? 0x80000000 : 0x7FFFFFFF; | 1783 | return aSign ? 0x80000000 : 0x7FFFFFFF; |
| 1793 | } | 1784 | } |
| 1794 | if ( ( aSig<<shiftCount ) != savedASig ) { | 1785 | if ( ( aSig<<shiftCount ) != savedASig ) { |
| 1795 | float_exception_flags |= float_flag_inexact; | 1786 | float_raise( float_flag_inexact ); |
| 1796 | } | 1787 | } |
| 1797 | return z; | 1788 | return z; |
| 1798 | } | 1789 | } |
| @@ -1805,7 +1796,7 @@ performed according to the IEC/IEEE Standard for Binary Floating-point | |||
| 1805 | Arithmetic. | 1796 | Arithmetic. |
| 1806 | ------------------------------------------------------------------------------- | 1797 | ------------------------------------------------------------------------------- |
| 1807 | */ | 1798 | */ |
| 1808 | float32 float64_to_float32( float64 a ) | 1799 | float32 float64_to_float32( struct roundingData *roundData, float64 a ) |
| 1809 | { | 1800 | { |
| 1810 | flag aSign; | 1801 | flag aSign; |
| 1811 | int16 aExp; | 1802 | int16 aExp; |
| @@ -1825,7 +1816,7 @@ float32 float64_to_float32( float64 a ) | |||
| 1825 | zSig |= 0x40000000; | 1816 | zSig |= 0x40000000; |
| 1826 | aExp -= 0x381; | 1817 | aExp -= 0x381; |
| 1827 | } | 1818 | } |
| 1828 | return roundAndPackFloat32( aSign, aExp, zSig ); | 1819 | return roundAndPackFloat32( roundData, aSign, aExp, zSig ); |
| 1829 | 1820 | ||
| 1830 | } | 1821 | } |
| 1831 | 1822 | ||
| @@ -1872,7 +1863,7 @@ operation is performed according to the IEC/IEEE Standard for Binary | |||
| 1872 | Floating-point Arithmetic. | 1863 | Floating-point Arithmetic. |
| 1873 | ------------------------------------------------------------------------------- | 1864 | ------------------------------------------------------------------------------- |
| 1874 | */ | 1865 | */ |
| 1875 | float64 float64_round_to_int( float64 a ) | 1866 | float64 float64_round_to_int( struct roundingData *roundData, float64 a ) |
| 1876 | { | 1867 | { |
| 1877 | flag aSign; | 1868 | flag aSign; |
| 1878 | int16 aExp; | 1869 | int16 aExp; |
| @@ -1889,9 +1880,9 @@ float64 float64_round_to_int( float64 a ) | |||
| 1889 | } | 1880 | } |
| 1890 | if ( aExp <= 0x3FE ) { | 1881 | if ( aExp <= 0x3FE ) { |
| 1891 | if ( (bits64) ( a<<1 ) == 0 ) return a; | 1882 | if ( (bits64) ( a<<1 ) == 0 ) return a; |
| 1892 | float_exception_flags |= float_flag_inexact; | 1883 | roundData->exception |= float_flag_inexact; |
| 1893 | aSign = extractFloat64Sign( a ); | 1884 | aSign = extractFloat64Sign( a ); |
| 1894 | switch ( float_rounding_mode ) { | 1885 | switch ( roundData->mode ) { |
| 1895 | case float_round_nearest_even: | 1886 | case float_round_nearest_even: |
| 1896 | if ( ( aExp == 0x3FE ) && extractFloat64Frac( a ) ) { | 1887 | if ( ( aExp == 0x3FE ) && extractFloat64Frac( a ) ) { |
| 1897 | return packFloat64( aSign, 0x3FF, 0 ); | 1888 | return packFloat64( aSign, 0x3FF, 0 ); |
| @@ -1909,7 +1900,7 @@ float64 float64_round_to_int( float64 a ) | |||
| 1909 | lastBitMask <<= 0x433 - aExp; | 1900 | lastBitMask <<= 0x433 - aExp; |
| 1910 | roundBitsMask = lastBitMask - 1; | 1901 | roundBitsMask = lastBitMask - 1; |
| 1911 | z = a; | 1902 | z = a; |
| 1912 | roundingMode = float_rounding_mode; | 1903 | roundingMode = roundData->mode; |
| 1913 | if ( roundingMode == float_round_nearest_even ) { | 1904 | if ( roundingMode == float_round_nearest_even ) { |
| 1914 | z += lastBitMask>>1; | 1905 | z += lastBitMask>>1; |
| 1915 | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; | 1906 | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; |
| @@ -1920,7 +1911,7 @@ float64 float64_round_to_int( float64 a ) | |||
| 1920 | } | 1911 | } |
| 1921 | } | 1912 | } |
| 1922 | z &= ~ roundBitsMask; | 1913 | z &= ~ roundBitsMask; |
| 1923 | if ( z != a ) float_exception_flags |= float_flag_inexact; | 1914 | if ( z != a ) roundData->exception |= float_flag_inexact; |
| 1924 | return z; | 1915 | return z; |
| 1925 | 1916 | ||
| 1926 | } | 1917 | } |
| @@ -1934,7 +1925,7 @@ addition is performed according to the IEC/IEEE Standard for Binary | |||
| 1934 | Floating-point Arithmetic. | 1925 | Floating-point Arithmetic. |
| 1935 | ------------------------------------------------------------------------------- | 1926 | ------------------------------------------------------------------------------- |
| 1936 | */ | 1927 | */ |
| 1937 | static float64 addFloat64Sigs( float64 a, float64 b, flag zSign ) | 1928 | static float64 addFloat64Sigs( struct roundingData *roundData, float64 a, float64 b, flag zSign ) |
| 1938 | { | 1929 | { |
| 1939 | int16 aExp, bExp, zExp; | 1930 | int16 aExp, bExp, zExp; |
| 1940 | bits64 aSig, bSig, zSig; | 1931 | bits64 aSig, bSig, zSig; |
| @@ -1993,7 +1984,7 @@ static float64 addFloat64Sigs( float64 a, float64 b, flag zSign ) | |||
| 1993 | ++zExp; | 1984 | ++zExp; |
| 1994 | } | 1985 | } |
| 1995 | roundAndPack: | 1986 | roundAndPack: |
| 1996 | return roundAndPackFloat64( zSign, zExp, zSig ); | 1987 | return roundAndPackFloat64( roundData, zSign, zExp, zSig ); |
| 1997 | 1988 | ||
| 1998 | } | 1989 | } |
| 1999 | 1990 | ||
| @@ -2006,7 +1997,7 @@ result is a NaN. The subtraction is performed according to the IEC/IEEE | |||
| 2006 | Standard for Binary Floating-point Arithmetic. | 1997 | Standard for Binary Floating-point Arithmetic. |
| 2007 | ------------------------------------------------------------------------------- | 1998 | ------------------------------------------------------------------------------- |
| 2008 | */ | 1999 | */ |
| 2009 | static float64 subFloat64Sigs( float64 a, float64 b, flag zSign ) | 2000 | static float64 subFloat64Sigs( struct roundingData *roundData, float64 a, float64 b, flag zSign ) |
| 2010 | { | 2001 | { |
| 2011 | int16 aExp, bExp, zExp; | 2002 | int16 aExp, bExp, zExp; |
| 2012 | bits64 aSig, bSig, zSig; | 2003 | bits64 aSig, bSig, zSig; |
| @@ -2023,7 +2014,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign ) | |||
| 2023 | if ( expDiff < 0 ) goto bExpBigger; | 2014 | if ( expDiff < 0 ) goto bExpBigger; |
| 2024 | if ( aExp == 0x7FF ) { | 2015 | if ( aExp == 0x7FF ) { |
| 2025 | if ( aSig | bSig ) return propagateFloat64NaN( a, b ); | 2016 | if ( aSig | bSig ) return propagateFloat64NaN( a, b ); |
| 2026 | float_raise( float_flag_invalid ); | 2017 | roundData->exception |= float_flag_invalid; |
| 2027 | return float64_default_nan; | 2018 | return float64_default_nan; |
| 2028 | } | 2019 | } |
| 2029 | if ( aExp == 0 ) { | 2020 | if ( aExp == 0 ) { |
| @@ -2032,7 +2023,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign ) | |||
| 2032 | } | 2023 | } |
| 2033 | if ( bSig < aSig ) goto aBigger; | 2024 | if ( bSig < aSig ) goto aBigger; |
| 2034 | if ( aSig < bSig ) goto bBigger; | 2025 | if ( aSig < bSig ) goto bBigger; |
| 2035 | return packFloat64( float_rounding_mode == float_round_down, 0, 0 ); | 2026 | return packFloat64( roundData->mode == float_round_down, 0, 0 ); |
| 2036 | bExpBigger: | 2027 | bExpBigger: |
| 2037 | if ( bExp == 0x7FF ) { | 2028 | if ( bExp == 0x7FF ) { |
| 2038 | if ( bSig ) return propagateFloat64NaN( a, b ); | 2029 | if ( bSig ) return propagateFloat64NaN( a, b ); |
| @@ -2069,7 +2060,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign ) | |||
| 2069 | zExp = aExp; | 2060 | zExp = aExp; |
| 2070 | normalizeRoundAndPack: | 2061 | normalizeRoundAndPack: |
| 2071 | --zExp; | 2062 | --zExp; |
| 2072 | return normalizeRoundAndPackFloat64( zSign, zExp, zSig ); | 2063 | return normalizeRoundAndPackFloat64( roundData, zSign, zExp, zSig ); |
| 2073 | 2064 | ||
| 2074 | } | 2065 | } |
| 2075 | 2066 | ||
| @@ -2080,17 +2071,17 @@ and `b'. The operation is performed according to the IEC/IEEE Standard for | |||
| 2080 | Binary Floating-point Arithmetic. | 2071 | Binary Floating-point Arithmetic. |
| 2081 | ------------------------------------------------------------------------------- | 2072 | ------------------------------------------------------------------------------- |
| 2082 | */ | 2073 | */ |
| 2083 | float64 float64_add( float64 a, float64 b ) | 2074 | float64 float64_add( struct roundingData *roundData, float64 a, float64 b ) |
| 2084 | { | 2075 | { |
| 2085 | flag aSign, bSign; | 2076 | flag aSign, bSign; |
| 2086 | 2077 | ||
| 2087 | aSign = extractFloat64Sign( a ); | 2078 | aSign = extractFloat64Sign( a ); |
| 2088 | bSign = extractFloat64Sign( b ); | 2079 | bSign = extractFloat64Sign( b ); |
| 2089 | if ( aSign == bSign ) { | 2080 | if ( aSign == bSign ) { |
| 2090 | return addFloat64Sigs( a, b, aSign ); | 2081 | return addFloat64Sigs( roundData, a, b, aSign ); |
| 2091 | } | 2082 | } |
| 2092 | else { | 2083 | else { |
| 2093 | return subFloat64Sigs( a, b, aSign ); | 2084 | return subFloat64Sigs( roundData, a, b, aSign ); |
| 2094 | } | 2085 | } |
| 2095 | 2086 | ||
| 2096 | } | 2087 | } |
| @@ -2102,17 +2093,17 @@ Returns the result of subtracting the double-precision floating-point values | |||
| 2102 | for Binary Floating-point Arithmetic. | 2093 | for Binary Floating-point Arithmetic. |
| 2103 | ------------------------------------------------------------------------------- | 2094 | ------------------------------------------------------------------------------- |
| 2104 | */ | 2095 | */ |
| 2105 | float64 float64_sub( float64 a, float64 b ) | 2096 | float64 float64_sub( struct roundingData *roundData, float64 a, float64 b ) |
| 2106 | { | 2097 | { |
| 2107 | flag aSign, bSign; | 2098 | flag aSign, bSign; |
| 2108 | 2099 | ||
| 2109 | aSign = extractFloat64Sign( a ); | 2100 | aSign = extractFloat64Sign( a ); |
| 2110 | bSign = extractFloat64Sign( b ); | 2101 | bSign = extractFloat64Sign( b ); |
| 2111 | if ( aSign == bSign ) { | 2102 | if ( aSign == bSign ) { |
| 2112 | return subFloat64Sigs( a, b, aSign ); | 2103 | return subFloat64Sigs( roundData, a, b, aSign ); |
| 2113 | } | 2104 | } |
| 2114 | else { | 2105 | else { |
| 2115 | return addFloat64Sigs( a, b, aSign ); | 2106 | return addFloat64Sigs( roundData, a, b, aSign ); |
| 2116 | } | 2107 | } |
| 2117 | 2108 | ||
| 2118 | } | 2109 | } |
| @@ -2124,7 +2115,7 @@ Returns the result of multiplying the double-precision floating-point values | |||
| 2124 | for Binary Floating-point Arithmetic. | 2115 | for Binary Floating-point Arithmetic. |
| 2125 | ------------------------------------------------------------------------------- | 2116 | ------------------------------------------------------------------------------- |
| 2126 | */ | 2117 | */ |
| 2127 | float64 float64_mul( float64 a, float64 b ) | 2118 | float64 float64_mul( struct roundingData *roundData, float64 a, float64 b ) |
| 2128 | { | 2119 | { |
| 2129 | flag aSign, bSign, zSign; | 2120 | flag aSign, bSign, zSign; |
| 2130 | int16 aExp, bExp, zExp; | 2121 | int16 aExp, bExp, zExp; |
| @@ -2142,7 +2133,7 @@ float64 float64_mul( float64 a, float64 b ) | |||
| 2142 | return propagateFloat64NaN( a, b ); | 2133 | return propagateFloat64NaN( a, b ); |
| 2143 | } | 2134 | } |
| 2144 | if ( ( bExp | bSig ) == 0 ) { | 2135 | if ( ( bExp | bSig ) == 0 ) { |
| 2145 | float_raise( float_flag_invalid ); | 2136 | roundData->exception |= float_flag_invalid; |
| 2146 | return float64_default_nan; | 2137 | return float64_default_nan; |
| 2147 | } | 2138 | } |
| 2148 | return packFloat64( zSign, 0x7FF, 0 ); | 2139 | return packFloat64( zSign, 0x7FF, 0 ); |
| @@ -2150,7 +2141,7 @@ float64 float64_mul( float64 a, float64 b ) | |||
| 2150 | if ( bExp == 0x7FF ) { | 2141 | if ( bExp == 0x7FF ) { |
| 2151 | if ( bSig ) return propagateFloat64NaN( a, b ); | 2142 | if ( bSig ) return propagateFloat64NaN( a, b ); |
| 2152 | if ( ( aExp | aSig ) == 0 ) { | 2143 | if ( ( aExp | aSig ) == 0 ) { |
| 2153 | float_raise( float_flag_invalid ); | 2144 | roundData->exception |= float_flag_invalid; |
| 2154 | return float64_default_nan; | 2145 | return float64_default_nan; |
| 2155 | } | 2146 | } |
| 2156 | return packFloat64( zSign, 0x7FF, 0 ); | 2147 | return packFloat64( zSign, 0x7FF, 0 ); |
| @@ -2172,7 +2163,7 @@ float64 float64_mul( float64 a, float64 b ) | |||
| 2172 | zSig0 <<= 1; | 2163 | zSig0 <<= 1; |
| 2173 | --zExp; | 2164 | --zExp; |
| 2174 | } | 2165 | } |
| 2175 | return roundAndPackFloat64( zSign, zExp, zSig0 ); | 2166 | return roundAndPackFloat64( roundData, zSign, zExp, zSig0 ); |
| 2176 | 2167 | ||
| 2177 | } | 2168 | } |
| 2178 | 2169 | ||
| @@ -2183,7 +2174,7 @@ by the corresponding value `b'. The operation is performed according to | |||
| 2183 | the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 2174 | the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 2184 | ------------------------------------------------------------------------------- | 2175 | ------------------------------------------------------------------------------- |
| 2185 | */ | 2176 | */ |
| 2186 | float64 float64_div( float64 a, float64 b ) | 2177 | float64 float64_div( struct roundingData *roundData, float64 a, float64 b ) |
| 2187 | { | 2178 | { |
| 2188 | flag aSign, bSign, zSign; | 2179 | flag aSign, bSign, zSign; |
| 2189 | int16 aExp, bExp, zExp; | 2180 | int16 aExp, bExp, zExp; |
| @@ -2202,7 +2193,7 @@ float64 float64_div( float64 a, float64 b ) | |||
| 2202 | if ( aSig ) return propagateFloat64NaN( a, b ); | 2193 | if ( aSig ) return propagateFloat64NaN( a, b ); |
| 2203 | if ( bExp == 0x7FF ) { | 2194 | if ( bExp == 0x7FF ) { |
| 2204 | if ( bSig ) return propagateFloat64NaN( a, b ); | 2195 | if ( bSig ) return propagateFloat64NaN( a, b ); |
| 2205 | float_raise( float_flag_invalid ); | 2196 | roundData->exception |= float_flag_invalid; |
| 2206 | return float64_default_nan; | 2197 | return float64_default_nan; |
| 2207 | } | 2198 | } |
| 2208 | return packFloat64( zSign, 0x7FF, 0 ); | 2199 | return packFloat64( zSign, 0x7FF, 0 ); |
| @@ -2214,10 +2205,10 @@ float64 float64_div( float64 a, float64 b ) | |||
| 2214 | if ( bExp == 0 ) { | 2205 | if ( bExp == 0 ) { |
| 2215 | if ( bSig == 0 ) { | 2206 | if ( bSig == 0 ) { |
| 2216 | if ( ( aExp | aSig ) == 0 ) { | 2207 | if ( ( aExp | aSig ) == 0 ) { |
| 2217 | float_raise( float_flag_invalid ); | 2208 | roundData->exception |= float_flag_invalid; |
| 2218 | return float64_default_nan; | 2209 | return float64_default_nan; |
| 2219 | } | 2210 | } |
| 2220 | float_raise( float_flag_divbyzero ); | 2211 | roundData->exception |= float_flag_divbyzero; |
| 2221 | return packFloat64( zSign, 0x7FF, 0 ); | 2212 | return packFloat64( zSign, 0x7FF, 0 ); |
| 2222 | } | 2213 | } |
| 2223 | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); | 2214 | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); |
| @@ -2243,7 +2234,7 @@ float64 float64_div( float64 a, float64 b ) | |||
| 2243 | } | 2234 | } |
| 2244 | zSig |= ( rem1 != 0 ); | 2235 | zSig |= ( rem1 != 0 ); |
| 2245 | } | 2236 | } |
| 2246 | return roundAndPackFloat64( zSign, zExp, zSig ); | 2237 | return roundAndPackFloat64( roundData, zSign, zExp, zSig ); |
| 2247 | 2238 | ||
| 2248 | } | 2239 | } |
| 2249 | 2240 | ||
| @@ -2254,7 +2245,7 @@ with respect to the corresponding value `b'. The operation is performed | |||
| 2254 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 2245 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 2255 | ------------------------------------------------------------------------------- | 2246 | ------------------------------------------------------------------------------- |
| 2256 | */ | 2247 | */ |
| 2257 | float64 float64_rem( float64 a, float64 b ) | 2248 | float64 float64_rem( struct roundingData *roundData, float64 a, float64 b ) |
| 2258 | { | 2249 | { |
| 2259 | flag aSign, bSign, zSign; | 2250 | flag aSign, bSign, zSign; |
| 2260 | int16 aExp, bExp, expDiff; | 2251 | int16 aExp, bExp, expDiff; |
| @@ -2272,7 +2263,7 @@ float64 float64_rem( float64 a, float64 b ) | |||
| 2272 | if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { | 2263 | if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { |
| 2273 | return propagateFloat64NaN( a, b ); | 2264 | return propagateFloat64NaN( a, b ); |
| 2274 | } | 2265 | } |
| 2275 | float_raise( float_flag_invalid ); | 2266 | roundData->exception |= float_flag_invalid; |
| 2276 | return float64_default_nan; | 2267 | return float64_default_nan; |
| 2277 | } | 2268 | } |
| 2278 | if ( bExp == 0x7FF ) { | 2269 | if ( bExp == 0x7FF ) { |
| @@ -2281,7 +2272,7 @@ float64 float64_rem( float64 a, float64 b ) | |||
| 2281 | } | 2272 | } |
| 2282 | if ( bExp == 0 ) { | 2273 | if ( bExp == 0 ) { |
| 2283 | if ( bSig == 0 ) { | 2274 | if ( bSig == 0 ) { |
| 2284 | float_raise( float_flag_invalid ); | 2275 | roundData->exception |= float_flag_invalid; |
| 2285 | return float64_default_nan; | 2276 | return float64_default_nan; |
| 2286 | } | 2277 | } |
| 2287 | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); | 2278 | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); |
| @@ -2329,7 +2320,7 @@ float64 float64_rem( float64 a, float64 b ) | |||
| 2329 | } | 2320 | } |
| 2330 | zSign = ( (sbits64) aSig < 0 ); | 2321 | zSign = ( (sbits64) aSig < 0 ); |
| 2331 | if ( zSign ) aSig = - aSig; | 2322 | if ( zSign ) aSig = - aSig; |
| 2332 | return normalizeRoundAndPackFloat64( aSign ^ zSign, bExp, aSig ); | 2323 | return normalizeRoundAndPackFloat64( roundData, aSign ^ zSign, bExp, aSig ); |
| 2333 | 2324 | ||
| 2334 | } | 2325 | } |
| 2335 | 2326 | ||
| @@ -2340,7 +2331,7 @@ The operation is performed according to the IEC/IEEE Standard for Binary | |||
| 2340 | Floating-point Arithmetic. | 2331 | Floating-point Arithmetic. |
| 2341 | ------------------------------------------------------------------------------- | 2332 | ------------------------------------------------------------------------------- |
| 2342 | */ | 2333 | */ |
| 2343 | float64 float64_sqrt( float64 a ) | 2334 | float64 float64_sqrt( struct roundingData *roundData, float64 a ) |
| 2344 | { | 2335 | { |
| 2345 | flag aSign; | 2336 | flag aSign; |
| 2346 | int16 aExp, zExp; | 2337 | int16 aExp, zExp; |
| @@ -2354,12 +2345,12 @@ float64 float64_sqrt( float64 a ) | |||
| 2354 | if ( aExp == 0x7FF ) { | 2345 | if ( aExp == 0x7FF ) { |
| 2355 | if ( aSig ) return propagateFloat64NaN( a, a ); | 2346 | if ( aSig ) return propagateFloat64NaN( a, a ); |
| 2356 | if ( ! aSign ) return a; | 2347 | if ( ! aSign ) return a; |
| 2357 | float_raise( float_flag_invalid ); | 2348 | roundData->exception |= float_flag_invalid; |
| 2358 | return float64_default_nan; | 2349 | return float64_default_nan; |
| 2359 | } | 2350 | } |
| 2360 | if ( aSign ) { | 2351 | if ( aSign ) { |
| 2361 | if ( ( aExp | aSig ) == 0 ) return a; | 2352 | if ( ( aExp | aSig ) == 0 ) return a; |
| 2362 | float_raise( float_flag_invalid ); | 2353 | roundData->exception |= float_flag_invalid; |
| 2363 | return float64_default_nan; | 2354 | return float64_default_nan; |
| 2364 | } | 2355 | } |
| 2365 | if ( aExp == 0 ) { | 2356 | if ( aExp == 0 ) { |
| @@ -2390,7 +2381,7 @@ float64 float64_sqrt( float64 a ) | |||
| 2390 | } | 2381 | } |
| 2391 | } | 2382 | } |
| 2392 | shift64RightJamming( zSig, 1, &zSig ); | 2383 | shift64RightJamming( zSig, 1, &zSig ); |
| 2393 | return roundAndPackFloat64( 0, zExp, zSig ); | 2384 | return roundAndPackFloat64( roundData, 0, zExp, zSig ); |
| 2394 | 2385 | ||
| 2395 | } | 2386 | } |
| 2396 | 2387 | ||
| @@ -2554,7 +2545,7 @@ largest positive integer is returned. Otherwise, if the conversion | |||
| 2554 | overflows, the largest integer with the same sign as `a' is returned. | 2545 | overflows, the largest integer with the same sign as `a' is returned. |
| 2555 | ------------------------------------------------------------------------------- | 2546 | ------------------------------------------------------------------------------- |
| 2556 | */ | 2547 | */ |
| 2557 | int32 floatx80_to_int32( floatx80 a ) | 2548 | int32 floatx80_to_int32( struct roundingData *roundData, floatx80 a ) |
| 2558 | { | 2549 | { |
| 2559 | flag aSign; | 2550 | flag aSign; |
| 2560 | int32 aExp, shiftCount; | 2551 | int32 aExp, shiftCount; |
| @@ -2567,7 +2558,7 @@ int32 floatx80_to_int32( floatx80 a ) | |||
| 2567 | shiftCount = 0x4037 - aExp; | 2558 | shiftCount = 0x4037 - aExp; |
| 2568 | if ( shiftCount <= 0 ) shiftCount = 1; | 2559 | if ( shiftCount <= 0 ) shiftCount = 1; |
| 2569 | shift64RightJamming( aSig, shiftCount, &aSig ); | 2560 | shift64RightJamming( aSig, shiftCount, &aSig ); |
| 2570 | return roundAndPackInt32( aSign, aSig ); | 2561 | return roundAndPackInt32( roundData, aSign, aSig ); |
| 2571 | 2562 | ||
| 2572 | } | 2563 | } |
| 2573 | 2564 | ||
| @@ -2598,7 +2589,7 @@ int32 floatx80_to_int32_round_to_zero( floatx80 a ) | |||
| 2598 | goto invalid; | 2589 | goto invalid; |
| 2599 | } | 2590 | } |
| 2600 | else if ( 63 < shiftCount ) { | 2591 | else if ( 63 < shiftCount ) { |
| 2601 | if ( aExp || aSig ) float_exception_flags |= float_flag_inexact; | 2592 | if ( aExp || aSig ) float_raise( float_flag_inexact ); |
| 2602 | return 0; | 2593 | return 0; |
| 2603 | } | 2594 | } |
| 2604 | savedASig = aSig; | 2595 | savedASig = aSig; |
| @@ -2607,11 +2598,11 @@ int32 floatx80_to_int32_round_to_zero( floatx80 a ) | |||
| 2607 | if ( aSign ) z = - z; | 2598 | if ( aSign ) z = - z; |
| 2608 | if ( ( z < 0 ) ^ aSign ) { | 2599 | if ( ( z < 0 ) ^ aSign ) { |
| 2609 | invalid: | 2600 | invalid: |
| 2610 | float_exception_flags |= float_flag_invalid; | 2601 | float_raise( float_flag_invalid ); |
| 2611 | return aSign ? 0x80000000 : 0x7FFFFFFF; | 2602 | return aSign ? 0x80000000 : 0x7FFFFFFF; |
| 2612 | } | 2603 | } |
| 2613 | if ( ( aSig<<shiftCount ) != savedASig ) { | 2604 | if ( ( aSig<<shiftCount ) != savedASig ) { |
| 2614 | float_exception_flags |= float_flag_inexact; | 2605 | float_raise( float_flag_inexact ); |
| 2615 | } | 2606 | } |
| 2616 | return z; | 2607 | return z; |
| 2617 | 2608 | ||
| @@ -2625,7 +2616,7 @@ conversion is performed according to the IEC/IEEE Standard for Binary | |||
| 2625 | Floating-point Arithmetic. | 2616 | Floating-point Arithmetic. |
| 2626 | ------------------------------------------------------------------------------- | 2617 | ------------------------------------------------------------------------------- |
| 2627 | */ | 2618 | */ |
| 2628 | float32 floatx80_to_float32( floatx80 a ) | 2619 | float32 floatx80_to_float32( struct roundingData *roundData, floatx80 a ) |
| 2629 | { | 2620 | { |
| 2630 | flag aSign; | 2621 | flag aSign; |
| 2631 | int32 aExp; | 2622 | int32 aExp; |
| @@ -2642,7 +2633,7 @@ float32 floatx80_to_float32( floatx80 a ) | |||
| 2642 | } | 2633 | } |
| 2643 | shift64RightJamming( aSig, 33, &aSig ); | 2634 | shift64RightJamming( aSig, 33, &aSig ); |
| 2644 | if ( aExp || aSig ) aExp -= 0x3F81; | 2635 | if ( aExp || aSig ) aExp -= 0x3F81; |
| 2645 | return roundAndPackFloat32( aSign, aExp, aSig ); | 2636 | return roundAndPackFloat32( roundData, aSign, aExp, aSig ); |
| 2646 | 2637 | ||
| 2647 | } | 2638 | } |
| 2648 | 2639 | ||
| @@ -2654,7 +2645,7 @@ conversion is performed according to the IEC/IEEE Standard for Binary | |||
| 2654 | Floating-point Arithmetic. | 2645 | Floating-point Arithmetic. |
| 2655 | ------------------------------------------------------------------------------- | 2646 | ------------------------------------------------------------------------------- |
| 2656 | */ | 2647 | */ |
| 2657 | float64 floatx80_to_float64( floatx80 a ) | 2648 | float64 floatx80_to_float64( struct roundingData *roundData, floatx80 a ) |
| 2658 | { | 2649 | { |
| 2659 | flag aSign; | 2650 | flag aSign; |
| 2660 | int32 aExp; | 2651 | int32 aExp; |
| @@ -2671,7 +2662,7 @@ float64 floatx80_to_float64( floatx80 a ) | |||
| 2671 | } | 2662 | } |
| 2672 | shift64RightJamming( aSig, 1, &zSig ); | 2663 | shift64RightJamming( aSig, 1, &zSig ); |
| 2673 | if ( aExp || aSig ) aExp -= 0x3C01; | 2664 | if ( aExp || aSig ) aExp -= 0x3C01; |
| 2674 | return roundAndPackFloat64( aSign, aExp, zSig ); | 2665 | return roundAndPackFloat64( roundData, aSign, aExp, zSig ); |
| 2675 | 2666 | ||
| 2676 | } | 2667 | } |
| 2677 | 2668 | ||
| @@ -2683,7 +2674,7 @@ value. The operation is performed according to the IEC/IEEE Standard for | |||
| 2683 | Binary Floating-point Arithmetic. | 2674 | Binary Floating-point Arithmetic. |
| 2684 | ------------------------------------------------------------------------------- | 2675 | ------------------------------------------------------------------------------- |
| 2685 | */ | 2676 | */ |
| 2686 | floatx80 floatx80_round_to_int( floatx80 a ) | 2677 | floatx80 floatx80_round_to_int( struct roundingData *roundData, floatx80 a ) |
| 2687 | { | 2678 | { |
| 2688 | flag aSign; | 2679 | flag aSign; |
| 2689 | int32 aExp; | 2680 | int32 aExp; |
| @@ -2703,9 +2694,9 @@ floatx80 floatx80_round_to_int( floatx80 a ) | |||
| 2703 | && ( (bits64) ( extractFloatx80Frac( a )<<1 ) == 0 ) ) { | 2694 | && ( (bits64) ( extractFloatx80Frac( a )<<1 ) == 0 ) ) { |
| 2704 | return a; | 2695 | return a; |
| 2705 | } | 2696 | } |
| 2706 | float_exception_flags |= float_flag_inexact; | 2697 | roundData->exception |= float_flag_inexact; |
| 2707 | aSign = extractFloatx80Sign( a ); | 2698 | aSign = extractFloatx80Sign( a ); |
| 2708 | switch ( float_rounding_mode ) { | 2699 | switch ( roundData->mode ) { |
| 2709 | case float_round_nearest_even: | 2700 | case float_round_nearest_even: |
| 2710 | if ( ( aExp == 0x3FFE ) && (bits64) ( extractFloatx80Frac( a )<<1 ) | 2701 | if ( ( aExp == 0x3FFE ) && (bits64) ( extractFloatx80Frac( a )<<1 ) |
| 2711 | ) { | 2702 | ) { |
| @@ -2729,7 +2720,7 @@ floatx80 floatx80_round_to_int( floatx80 a ) | |||
| 2729 | lastBitMask <<= 0x403E - aExp; | 2720 | lastBitMask <<= 0x403E - aExp; |
| 2730 | roundBitsMask = lastBitMask - 1; | 2721 | roundBitsMask = lastBitMask - 1; |
| 2731 | z = a; | 2722 | z = a; |
| 2732 | roundingMode = float_rounding_mode; | 2723 | roundingMode = roundData->mode; |
| 2733 | if ( roundingMode == float_round_nearest_even ) { | 2724 | if ( roundingMode == float_round_nearest_even ) { |
| 2734 | z.low += lastBitMask>>1; | 2725 | z.low += lastBitMask>>1; |
| 2735 | if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; | 2726 | if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; |
| @@ -2744,7 +2735,7 @@ floatx80 floatx80_round_to_int( floatx80 a ) | |||
| 2744 | ++z.high; | 2735 | ++z.high; |
| 2745 | z.low = LIT64( 0x8000000000000000 ); | 2736 | z.low = LIT64( 0x8000000000000000 ); |
| 2746 | } | 2737 | } |
| 2747 | if ( z.low != a.low ) float_exception_flags |= float_flag_inexact; | 2738 | if ( z.low != a.low ) roundData->exception |= float_flag_inexact; |
| 2748 | return z; | 2739 | return z; |
| 2749 | 2740 | ||
| 2750 | } | 2741 | } |
| @@ -2758,7 +2749,7 @@ The addition is performed according to the IEC/IEEE Standard for Binary | |||
| 2758 | Floating-point Arithmetic. | 2749 | Floating-point Arithmetic. |
| 2759 | ------------------------------------------------------------------------------- | 2750 | ------------------------------------------------------------------------------- |
| 2760 | */ | 2751 | */ |
| 2761 | static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | 2752 | static floatx80 addFloatx80Sigs( struct roundingData *roundData, floatx80 a, floatx80 b, flag zSign ) |
| 2762 | { | 2753 | { |
| 2763 | int32 aExp, bExp, zExp; | 2754 | int32 aExp, bExp, zExp; |
| 2764 | bits64 aSig, bSig, zSig0, zSig1; | 2755 | bits64 aSig, bSig, zSig0, zSig1; |
| @@ -2814,7 +2805,7 @@ static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | |||
| 2814 | roundAndPack: | 2805 | roundAndPack: |
| 2815 | return | 2806 | return |
| 2816 | roundAndPackFloatx80( | 2807 | roundAndPackFloatx80( |
| 2817 | floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); | 2808 | roundData, zSign, zExp, zSig0, zSig1 ); |
| 2818 | 2809 | ||
| 2819 | } | 2810 | } |
| 2820 | 2811 | ||
| @@ -2827,7 +2818,7 @@ result is a NaN. The subtraction is performed according to the IEC/IEEE | |||
| 2827 | Standard for Binary Floating-point Arithmetic. | 2818 | Standard for Binary Floating-point Arithmetic. |
| 2828 | ------------------------------------------------------------------------------- | 2819 | ------------------------------------------------------------------------------- |
| 2829 | */ | 2820 | */ |
| 2830 | static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | 2821 | static floatx80 subFloatx80Sigs( struct roundingData *roundData, floatx80 a, floatx80 b, flag zSign ) |
| 2831 | { | 2822 | { |
| 2832 | int32 aExp, bExp, zExp; | 2823 | int32 aExp, bExp, zExp; |
| 2833 | bits64 aSig, bSig, zSig0, zSig1; | 2824 | bits64 aSig, bSig, zSig0, zSig1; |
| @@ -2845,7 +2836,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | |||
| 2845 | if ( (bits64) ( ( aSig | bSig )<<1 ) ) { | 2836 | if ( (bits64) ( ( aSig | bSig )<<1 ) ) { |
| 2846 | return propagateFloatx80NaN( a, b ); | 2837 | return propagateFloatx80NaN( a, b ); |
| 2847 | } | 2838 | } |
| 2848 | float_raise( float_flag_invalid ); | 2839 | roundData->exception |= float_flag_invalid; |
| 2849 | z.low = floatx80_default_nan_low; | 2840 | z.low = floatx80_default_nan_low; |
| 2850 | z.high = floatx80_default_nan_high; | 2841 | z.high = floatx80_default_nan_high; |
| 2851 | return z; | 2842 | return z; |
| @@ -2857,7 +2848,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | |||
| 2857 | zSig1 = 0; | 2848 | zSig1 = 0; |
| 2858 | if ( bSig < aSig ) goto aBigger; | 2849 | if ( bSig < aSig ) goto aBigger; |
| 2859 | if ( aSig < bSig ) goto bBigger; | 2850 | if ( aSig < bSig ) goto bBigger; |
| 2860 | return packFloatx80( float_rounding_mode == float_round_down, 0, 0 ); | 2851 | return packFloatx80( roundData->mode == float_round_down, 0, 0 ); |
| 2861 | bExpBigger: | 2852 | bExpBigger: |
| 2862 | if ( bExp == 0x7FFF ) { | 2853 | if ( bExp == 0x7FFF ) { |
| 2863 | if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); | 2854 | if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); |
| @@ -2883,7 +2874,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) | |||
| 2883 | normalizeRoundAndPack: | 2874 | normalizeRoundAndPack: |
| 2884 | return | 2875 | return |
| 2885 | normalizeRoundAndPackFloatx80( | 2876 | normalizeRoundAndPackFloatx80( |
| 2886 | floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); | 2877 | roundData, zSign, zExp, zSig0, zSig1 ); |
| 2887 | 2878 | ||
| 2888 | } | 2879 | } |
| 2889 | 2880 | ||
| @@ -2894,17 +2885,17 @@ values `a' and `b'. The operation is performed according to the IEC/IEEE | |||
| 2894 | Standard for Binary Floating-point Arithmetic. | 2885 | Standard for Binary Floating-point Arithmetic. |
| 2895 | ------------------------------------------------------------------------------- | 2886 | ------------------------------------------------------------------------------- |
| 2896 | */ | 2887 | */ |
| 2897 | floatx80 floatx80_add( floatx80 a, floatx80 b ) | 2888 | floatx80 floatx80_add( struct roundingData *roundData, floatx80 a, floatx80 b ) |
| 2898 | { | 2889 | { |
| 2899 | flag aSign, bSign; | 2890 | flag aSign, bSign; |
| 2900 | 2891 | ||
| 2901 | aSign = extractFloatx80Sign( a ); | 2892 | aSign = extractFloatx80Sign( a ); |
| 2902 | bSign = extractFloatx80Sign( b ); | 2893 | bSign = extractFloatx80Sign( b ); |
| 2903 | if ( aSign == bSign ) { | 2894 | if ( aSign == bSign ) { |
| 2904 | return addFloatx80Sigs( a, b, aSign ); | 2895 | return addFloatx80Sigs( roundData, a, b, aSign ); |
| 2905 | } | 2896 | } |
| 2906 | else { | 2897 | else { |
| 2907 | return subFloatx80Sigs( a, b, aSign ); | 2898 | return subFloatx80Sigs( roundData, a, b, aSign ); |
| 2908 | } | 2899 | } |
| 2909 | 2900 | ||
| 2910 | } | 2901 | } |
| @@ -2916,17 +2907,17 @@ point values `a' and `b'. The operation is performed according to the | |||
| 2916 | IEC/IEEE Standard for Binary Floating-point Arithmetic. | 2907 | IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 2917 | ------------------------------------------------------------------------------- | 2908 | ------------------------------------------------------------------------------- |
| 2918 | */ | 2909 | */ |
| 2919 | floatx80 floatx80_sub( floatx80 a, floatx80 b ) | 2910 | floatx80 floatx80_sub( struct roundingData *roundData, floatx80 a, floatx80 b ) |
| 2920 | { | 2911 | { |
| 2921 | flag aSign, bSign; | 2912 | flag aSign, bSign; |
| 2922 | 2913 | ||
| 2923 | aSign = extractFloatx80Sign( a ); | 2914 | aSign = extractFloatx80Sign( a ); |
| 2924 | bSign = extractFloatx80Sign( b ); | 2915 | bSign = extractFloatx80Sign( b ); |
| 2925 | if ( aSign == bSign ) { | 2916 | if ( aSign == bSign ) { |
| 2926 | return subFloatx80Sigs( a, b, aSign ); | 2917 | return subFloatx80Sigs( roundData, a, b, aSign ); |
| 2927 | } | 2918 | } |
| 2928 | else { | 2919 | else { |
| 2929 | return addFloatx80Sigs( a, b, aSign ); | 2920 | return addFloatx80Sigs( roundData, a, b, aSign ); |
| 2930 | } | 2921 | } |
| 2931 | 2922 | ||
| 2932 | } | 2923 | } |
| @@ -2938,7 +2929,7 @@ point values `a' and `b'. The operation is performed according to the | |||
| 2938 | IEC/IEEE Standard for Binary Floating-point Arithmetic. | 2929 | IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 2939 | ------------------------------------------------------------------------------- | 2930 | ------------------------------------------------------------------------------- |
| 2940 | */ | 2931 | */ |
| 2941 | floatx80 floatx80_mul( floatx80 a, floatx80 b ) | 2932 | floatx80 floatx80_mul( struct roundingData *roundData, floatx80 a, floatx80 b ) |
| 2942 | { | 2933 | { |
| 2943 | flag aSign, bSign, zSign; | 2934 | flag aSign, bSign, zSign; |
| 2944 | int32 aExp, bExp, zExp; | 2935 | int32 aExp, bExp, zExp; |
| @@ -2964,7 +2955,7 @@ floatx80 floatx80_mul( floatx80 a, floatx80 b ) | |||
| 2964 | if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); | 2955 | if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); |
| 2965 | if ( ( aExp | aSig ) == 0 ) { | 2956 | if ( ( aExp | aSig ) == 0 ) { |
| 2966 | invalid: | 2957 | invalid: |
| 2967 | float_raise( float_flag_invalid ); | 2958 | roundData->exception |= float_flag_invalid; |
| 2968 | z.low = floatx80_default_nan_low; | 2959 | z.low = floatx80_default_nan_low; |
| 2969 | z.high = floatx80_default_nan_high; | 2960 | z.high = floatx80_default_nan_high; |
| 2970 | return z; | 2961 | return z; |
| @@ -2987,7 +2978,7 @@ floatx80 floatx80_mul( floatx80 a, floatx80 b ) | |||
| 2987 | } | 2978 | } |
| 2988 | return | 2979 | return |
| 2989 | roundAndPackFloatx80( | 2980 | roundAndPackFloatx80( |
| 2990 | floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); | 2981 | roundData, zSign, zExp, zSig0, zSig1 ); |
| 2991 | 2982 | ||
| 2992 | } | 2983 | } |
| 2993 | 2984 | ||
| @@ -2998,7 +2989,7 @@ value `a' by the corresponding value `b'. The operation is performed | |||
| 2998 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 2989 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 2999 | ------------------------------------------------------------------------------- | 2990 | ------------------------------------------------------------------------------- |
| 3000 | */ | 2991 | */ |
| 3001 | floatx80 floatx80_div( floatx80 a, floatx80 b ) | 2992 | floatx80 floatx80_div( struct roundingData *roundData, floatx80 a, floatx80 b ) |
| 3002 | { | 2993 | { |
| 3003 | flag aSign, bSign, zSign; | 2994 | flag aSign, bSign, zSign; |
| 3004 | int32 aExp, bExp, zExp; | 2995 | int32 aExp, bExp, zExp; |
| @@ -3029,12 +3020,12 @@ floatx80 floatx80_div( floatx80 a, floatx80 b ) | |||
| 3029 | if ( bSig == 0 ) { | 3020 | if ( bSig == 0 ) { |
| 3030 | if ( ( aExp | aSig ) == 0 ) { | 3021 | if ( ( aExp | aSig ) == 0 ) { |
| 3031 | invalid: | 3022 | invalid: |
| 3032 | float_raise( float_flag_invalid ); | 3023 | roundData->exception |= float_flag_invalid; |
| 3033 | z.low = floatx80_default_nan_low; | 3024 | z.low = floatx80_default_nan_low; |
| 3034 | z.high = floatx80_default_nan_high; | 3025 | z.high = floatx80_default_nan_high; |
| 3035 | return z; | 3026 | return z; |
| 3036 | } | 3027 | } |
| 3037 | float_raise( float_flag_divbyzero ); | 3028 | roundData->exception |= float_flag_divbyzero; |
| 3038 | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); | 3029 | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
| 3039 | } | 3030 | } |
| 3040 | normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); | 3031 | normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); |
| @@ -3068,7 +3059,7 @@ floatx80 floatx80_div( floatx80 a, floatx80 b ) | |||
| 3068 | } | 3059 | } |
| 3069 | return | 3060 | return |
| 3070 | roundAndPackFloatx80( | 3061 | roundAndPackFloatx80( |
| 3071 | floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); | 3062 | roundData, zSign, zExp, zSig0, zSig1 ); |
| 3072 | 3063 | ||
| 3073 | } | 3064 | } |
| 3074 | 3065 | ||
| @@ -3079,7 +3070,7 @@ Returns the remainder of the extended double-precision floating-point value | |||
| 3079 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. | 3070 | according to the IEC/IEEE Standard for Binary Floating-point Arithmetic. |
| 3080 | ------------------------------------------------------------------------------- | 3071 | ------------------------------------------------------------------------------- |
| 3081 | */ | 3072 | */ |
| 3082 | floatx80 floatx80_rem( floatx80 a, floatx80 b ) | 3073 | floatx80 floatx80_rem( struct roundingData *roundData, floatx80 a, floatx80 b ) |
| 3083 | { | 3074 | { |
| 3084 | flag aSign, bSign, zSign; | 3075 | flag aSign, bSign, zSign; |
| 3085 | int32 aExp, bExp, expDiff; | 3076 | int32 aExp, bExp, expDiff; |
| @@ -3107,7 +3098,7 @@ floatx80 floatx80_rem( floatx80 a, floatx80 b ) | |||
| 3107 | if ( bExp == 0 ) { | 3098 | if ( bExp == 0 ) { |
| 3108 | if ( bSig == 0 ) { | 3099 | if ( bSig == 0 ) { |
| 3109 | invalid: | 3100 | invalid: |
| 3110 | float_raise( float_flag_invalid ); | 3101 | roundData->exception |= float_flag_invalid; |
| 3111 | z.low = floatx80_default_nan_low; | 3102 | z.low = floatx80_default_nan_low; |
| 3112 | z.high = floatx80_default_nan_high; | 3103 | z.high = floatx80_default_nan_high; |
| 3113 | return z; | 3104 | return z; |
| @@ -3164,9 +3155,10 @@ floatx80 floatx80_rem( floatx80 a, floatx80 b ) | |||
| 3164 | aSig1 = alternateASig1; | 3155 | aSig1 = alternateASig1; |
| 3165 | zSign = ! zSign; | 3156 | zSign = ! zSign; |
| 3166 | } | 3157 | } |
| 3158 | |||
| 3167 | return | 3159 | return |
| 3168 | normalizeRoundAndPackFloatx80( | 3160 | normalizeRoundAndPackFloatx80( |
| 3169 | 80, zSign, bExp + expDiff, aSig0, aSig1 ); | 3161 | roundData, zSign, bExp + expDiff, aSig0, aSig1 ); |
| 3170 | 3162 | ||
| 3171 | } | 3163 | } |
| 3172 | 3164 | ||
| @@ -3177,7 +3169,7 @@ value `a'. The operation is performed according to the IEC/IEEE Standard | |||
| 3177 | for Binary Floating-point Arithmetic. | 3169 | for Binary Floating-point Arithmetic. |
| 3178 | ------------------------------------------------------------------------------- | 3170 | ------------------------------------------------------------------------------- |
| 3179 | */ | 3171 | */ |
| 3180 | floatx80 floatx80_sqrt( floatx80 a ) | 3172 | floatx80 floatx80_sqrt( struct roundingData *roundData, floatx80 a ) |
| 3181 | { | 3173 | { |
| 3182 | flag aSign; | 3174 | flag aSign; |
| 3183 | int32 aExp, zExp; | 3175 | int32 aExp, zExp; |
| @@ -3197,7 +3189,7 @@ floatx80 floatx80_sqrt( floatx80 a ) | |||
| 3197 | if ( aSign ) { | 3189 | if ( aSign ) { |
| 3198 | if ( ( aExp | aSig0 ) == 0 ) return a; | 3190 | if ( ( aExp | aSig0 ) == 0 ) return a; |
| 3199 | invalid: | 3191 | invalid: |
| 3200 | float_raise( float_flag_invalid ); | 3192 | roundData->exception |= float_flag_invalid; |
| 3201 | z.low = floatx80_default_nan_low; | 3193 | z.low = floatx80_default_nan_low; |
| 3202 | z.high = floatx80_default_nan_high; | 3194 | z.high = floatx80_default_nan_high; |
| 3203 | return z; | 3195 | return z; |
| @@ -3242,7 +3234,7 @@ floatx80 floatx80_sqrt( floatx80 a ) | |||
| 3242 | } | 3234 | } |
| 3243 | return | 3235 | return |
| 3244 | roundAndPackFloatx80( | 3236 | roundAndPackFloatx80( |
| 3245 | floatx80_rounding_precision, 0, zExp, zSig0, zSig1 ); | 3237 | roundData, 0, zExp, zSig0, zSig1 ); |
| 3246 | 3238 | ||
| 3247 | } | 3239 | } |
| 3248 | 3240 | ||
| @@ -3264,7 +3256,7 @@ flag floatx80_eq( floatx80 a, floatx80 b ) | |||
| 3264 | ) { | 3256 | ) { |
| 3265 | if ( floatx80_is_signaling_nan( a ) | 3257 | if ( floatx80_is_signaling_nan( a ) |
| 3266 | || floatx80_is_signaling_nan( b ) ) { | 3258 | || floatx80_is_signaling_nan( b ) ) { |
| 3267 | float_raise( float_flag_invalid ); | 3259 | roundData->exception |= float_flag_invalid; |
| 3268 | } | 3260 | } |
| 3269 | return 0; | 3261 | return 0; |
| 3270 | } | 3262 | } |
| @@ -3294,7 +3286,7 @@ flag floatx80_le( floatx80 a, floatx80 b ) | |||
| 3294 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) | 3286 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) |
| 3295 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) | 3287 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) |
| 3296 | ) { | 3288 | ) { |
| 3297 | float_raise( float_flag_invalid ); | 3289 | roundData->exception |= float_flag_invalid; |
| 3298 | return 0; | 3290 | return 0; |
| 3299 | } | 3291 | } |
| 3300 | aSign = extractFloatx80Sign( a ); | 3292 | aSign = extractFloatx80Sign( a ); |
| @@ -3328,7 +3320,7 @@ flag floatx80_lt( floatx80 a, floatx80 b ) | |||
| 3328 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) | 3320 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) |
| 3329 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) | 3321 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) |
| 3330 | ) { | 3322 | ) { |
| 3331 | float_raise( float_flag_invalid ); | 3323 | roundData->exception |= float_flag_invalid; |
| 3332 | return 0; | 3324 | return 0; |
| 3333 | } | 3325 | } |
| 3334 | aSign = extractFloatx80Sign( a ); | 3326 | aSign = extractFloatx80Sign( a ); |
| @@ -3361,7 +3353,7 @@ flag floatx80_eq_signaling( floatx80 a, floatx80 b ) | |||
| 3361 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) | 3353 | || ( ( extractFloatx80Exp( b ) == 0x7FFF ) |
| 3362 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) | 3354 | && (bits64) ( extractFloatx80Frac( b )<<1 ) ) |
| 3363 | ) { | 3355 | ) { |
| 3364 | float_raise( float_flag_invalid ); | 3356 | roundData->exception |= float_flag_invalid; |
| 3365 | return 0; | 3357 | return 0; |
| 3366 | } | 3358 | } |
| 3367 | return | 3359 | return |
| @@ -3392,7 +3384,7 @@ flag floatx80_le_quiet( floatx80 a, floatx80 b ) | |||
| 3392 | ) { | 3384 | ) { |
| 3393 | if ( floatx80_is_signaling_nan( a ) | 3385 | if ( floatx80_is_signaling_nan( a ) |
| 3394 | || floatx80_is_signaling_nan( b ) ) { | 3386 | || floatx80_is_signaling_nan( b ) ) { |
| 3395 | float_raise( float_flag_invalid ); | 3387 | roundData->exception |= float_flag_invalid; |
| 3396 | } | 3388 | } |
| 3397 | return 0; | 3389 | return 0; |
| 3398 | } | 3390 | } |
| @@ -3429,7 +3421,7 @@ flag floatx80_lt_quiet( floatx80 a, floatx80 b ) | |||
| 3429 | ) { | 3421 | ) { |
| 3430 | if ( floatx80_is_signaling_nan( a ) | 3422 | if ( floatx80_is_signaling_nan( a ) |
| 3431 | || floatx80_is_signaling_nan( b ) ) { | 3423 | || floatx80_is_signaling_nan( b ) ) { |
| 3432 | float_raise( float_flag_invalid ); | 3424 | roundData->exception |= float_flag_invalid; |
| 3433 | } | 3425 | } |
| 3434 | return 0; | 3426 | return 0; |
| 3435 | } | 3427 | } |
diff --git a/arch/arm/nwfpe/softfloat.h b/arch/arm/nwfpe/softfloat.h index 1e1743173899..1c8799b9ee4d 100644 --- a/arch/arm/nwfpe/softfloat.h +++ b/arch/arm/nwfpe/softfloat.h | |||
| @@ -74,7 +74,7 @@ enum { | |||
| 74 | Software IEC/IEEE floating-point rounding mode. | 74 | Software IEC/IEEE floating-point rounding mode. |
| 75 | ------------------------------------------------------------------------------- | 75 | ------------------------------------------------------------------------------- |
| 76 | */ | 76 | */ |
| 77 | extern signed char float_rounding_mode; | 77 | //extern int8 float_rounding_mode; |
| 78 | enum { | 78 | enum { |
| 79 | float_round_nearest_even = 0, | 79 | float_round_nearest_even = 0, |
| 80 | float_round_to_zero = 1, | 80 | float_round_to_zero = 1, |
| @@ -86,7 +86,6 @@ enum { | |||
| 86 | ------------------------------------------------------------------------------- | 86 | ------------------------------------------------------------------------------- |
| 87 | Software IEC/IEEE floating-point exception flags. | 87 | Software IEC/IEEE floating-point exception flags. |
| 88 | ------------------------------------------------------------------------------- | 88 | ------------------------------------------------------------------------------- |
| 89 | extern signed char float_exception_flags; | ||
| 90 | enum { | 89 | enum { |
| 91 | float_flag_inexact = 1, | 90 | float_flag_inexact = 1, |
| 92 | float_flag_underflow = 2, | 91 | float_flag_underflow = 2, |
| @@ -99,7 +98,6 @@ ScottB: November 4, 1998 | |||
| 99 | Changed the enumeration to match the bit order in the FPA11. | 98 | Changed the enumeration to match the bit order in the FPA11. |
| 100 | */ | 99 | */ |
| 101 | 100 | ||
| 102 | extern signed char float_exception_flags; | ||
| 103 | enum { | 101 | enum { |
| 104 | float_flag_invalid = 1, | 102 | float_flag_invalid = 1, |
| 105 | float_flag_divbyzero = 2, | 103 | float_flag_divbyzero = 2, |
| @@ -121,7 +119,7 @@ void float_raise( signed char ); | |||
| 121 | Software IEC/IEEE integer-to-floating-point conversion routines. | 119 | Software IEC/IEEE integer-to-floating-point conversion routines. |
| 122 | ------------------------------------------------------------------------------- | 120 | ------------------------------------------------------------------------------- |
| 123 | */ | 121 | */ |
| 124 | float32 int32_to_float32( signed int ); | 122 | float32 int32_to_float32( struct roundingData *, signed int ); |
| 125 | float64 int32_to_float64( signed int ); | 123 | float64 int32_to_float64( signed int ); |
| 126 | #ifdef FLOATX80 | 124 | #ifdef FLOATX80 |
| 127 | floatx80 int32_to_floatx80( signed int ); | 125 | floatx80 int32_to_floatx80( signed int ); |
| @@ -132,7 +130,7 @@ floatx80 int32_to_floatx80( signed int ); | |||
| 132 | Software IEC/IEEE single-precision conversion routines. | 130 | Software IEC/IEEE single-precision conversion routines. |
| 133 | ------------------------------------------------------------------------------- | 131 | ------------------------------------------------------------------------------- |
| 134 | */ | 132 | */ |
| 135 | signed int float32_to_int32( float32 ); | 133 | signed int float32_to_int32( struct roundingData *, float32 ); |
| 136 | signed int float32_to_int32_round_to_zero( float32 ); | 134 | signed int float32_to_int32_round_to_zero( float32 ); |
| 137 | float64 float32_to_float64( float32 ); | 135 | float64 float32_to_float64( float32 ); |
| 138 | #ifdef FLOATX80 | 136 | #ifdef FLOATX80 |
| @@ -144,13 +142,13 @@ floatx80 float32_to_floatx80( float32 ); | |||
| 144 | Software IEC/IEEE single-precision operations. | 142 | Software IEC/IEEE single-precision operations. |
| 145 | ------------------------------------------------------------------------------- | 143 | ------------------------------------------------------------------------------- |
| 146 | */ | 144 | */ |
| 147 | float32 float32_round_to_int( float32 ); | 145 | float32 float32_round_to_int( struct roundingData*, float32 ); |
| 148 | float32 float32_add( float32, float32 ); | 146 | float32 float32_add( struct roundingData *, float32, float32 ); |
| 149 | float32 float32_sub( float32, float32 ); | 147 | float32 float32_sub( struct roundingData *, float32, float32 ); |
| 150 | float32 float32_mul( float32, float32 ); | 148 | float32 float32_mul( struct roundingData *, float32, float32 ); |
| 151 | float32 float32_div( float32, float32 ); | 149 | float32 float32_div( struct roundingData *, float32, float32 ); |
| 152 | float32 float32_rem( float32, float32 ); | 150 | float32 float32_rem( struct roundingData *, float32, float32 ); |
| 153 | float32 float32_sqrt( float32 ); | 151 | float32 float32_sqrt( struct roundingData*, float32 ); |
| 154 | char float32_eq( float32, float32 ); | 152 | char float32_eq( float32, float32 ); |
| 155 | char float32_le( float32, float32 ); | 153 | char float32_le( float32, float32 ); |
| 156 | char float32_lt( float32, float32 ); | 154 | char float32_lt( float32, float32 ); |
| @@ -164,9 +162,9 @@ char float32_is_signaling_nan( float32 ); | |||
| 164 | Software IEC/IEEE double-precision conversion routines. | 162 | Software IEC/IEEE double-precision conversion routines. |
| 165 | ------------------------------------------------------------------------------- | 163 | ------------------------------------------------------------------------------- |
| 166 | */ | 164 | */ |
| 167 | signed int float64_to_int32( float64 ); | 165 | signed int float64_to_int32( struct roundingData *, float64 ); |
| 168 | signed int float64_to_int32_round_to_zero( float64 ); | 166 | signed int float64_to_int32_round_to_zero( float64 ); |
| 169 | float32 float64_to_float32( float64 ); | 167 | float32 float64_to_float32( struct roundingData *, float64 ); |
| 170 | #ifdef FLOATX80 | 168 | #ifdef FLOATX80 |
| 171 | floatx80 float64_to_floatx80( float64 ); | 169 | floatx80 float64_to_floatx80( float64 ); |
| 172 | #endif | 170 | #endif |
| @@ -176,13 +174,13 @@ floatx80 float64_to_floatx80( float64 ); | |||
| 176 | Software IEC/IEEE double-precision operations. | 174 | Software IEC/IEEE double-precision operations. |
| 177 | ------------------------------------------------------------------------------- | 175 | ------------------------------------------------------------------------------- |
| 178 | */ | 176 | */ |
| 179 | float64 float64_round_to_int( float64 ); | 177 | float64 float64_round_to_int( struct roundingData *, float64 ); |
| 180 | float64 float64_add( float64, float64 ); | 178 | float64 float64_add( struct roundingData *, float64, float64 ); |
| 181 | float64 float64_sub( float64, float64 ); | 179 | float64 float64_sub( struct roundingData *, float64, float64 ); |
| 182 | float64 float64_mul( float64, float64 ); | 180 | float64 float64_mul( struct roundingData *, float64, float64 ); |
| 183 | float64 float64_div( float64, float64 ); | 181 | float64 float64_div( struct roundingData *, float64, float64 ); |
| 184 | float64 float64_rem( float64, float64 ); | 182 | float64 float64_rem( struct roundingData *, float64, float64 ); |
| 185 | float64 float64_sqrt( float64 ); | 183 | float64 float64_sqrt( struct roundingData *, float64 ); |
| 186 | char float64_eq( float64, float64 ); | 184 | char float64_eq( float64, float64 ); |
| 187 | char float64_le( float64, float64 ); | 185 | char float64_le( float64, float64 ); |
| 188 | char float64_lt( float64, float64 ); | 186 | char float64_lt( float64, float64 ); |
| @@ -198,31 +196,23 @@ char float64_is_signaling_nan( float64 ); | |||
| 198 | Software IEC/IEEE extended double-precision conversion routines. | 196 | Software IEC/IEEE extended double-precision conversion routines. |
| 199 | ------------------------------------------------------------------------------- | 197 | ------------------------------------------------------------------------------- |
| 200 | */ | 198 | */ |
| 201 | signed int floatx80_to_int32( floatx80 ); | 199 | signed int floatx80_to_int32( struct roundingData *, floatx80 ); |
| 202 | signed int floatx80_to_int32_round_to_zero( floatx80 ); | 200 | signed int floatx80_to_int32_round_to_zero( floatx80 ); |
| 203 | float32 floatx80_to_float32( floatx80 ); | 201 | float32 floatx80_to_float32( struct roundingData *, floatx80 ); |
| 204 | float64 floatx80_to_float64( floatx80 ); | 202 | float64 floatx80_to_float64( struct roundingData *, floatx80 ); |
| 205 | |||
| 206 | /* | ||
| 207 | ------------------------------------------------------------------------------- | ||
| 208 | Software IEC/IEEE extended double-precision rounding precision. Valid | ||
| 209 | values are 32, 64, and 80. | ||
| 210 | ------------------------------------------------------------------------------- | ||
| 211 | */ | ||
| 212 | extern signed char floatx80_rounding_precision; | ||
| 213 | 203 | ||
| 214 | /* | 204 | /* |
| 215 | ------------------------------------------------------------------------------- | 205 | ------------------------------------------------------------------------------- |
| 216 | Software IEC/IEEE extended double-precision operations. | 206 | Software IEC/IEEE extended double-precision operations. |
| 217 | ------------------------------------------------------------------------------- | 207 | ------------------------------------------------------------------------------- |
| 218 | */ | 208 | */ |
| 219 | floatx80 floatx80_round_to_int( floatx80 ); | 209 | floatx80 floatx80_round_to_int( struct roundingData *, floatx80 ); |
| 220 | floatx80 floatx80_add( floatx80, floatx80 ); | 210 | floatx80 floatx80_add( struct roundingData *, floatx80, floatx80 ); |
| 221 | floatx80 floatx80_sub( floatx80, floatx80 ); | 211 | floatx80 floatx80_sub( struct roundingData *, floatx80, floatx80 ); |
| 222 | floatx80 floatx80_mul( floatx80, floatx80 ); | 212 | floatx80 floatx80_mul( struct roundingData *, floatx80, floatx80 ); |
| 223 | floatx80 floatx80_div( floatx80, floatx80 ); | 213 | floatx80 floatx80_div( struct roundingData *, floatx80, floatx80 ); |
| 224 | floatx80 floatx80_rem( floatx80, floatx80 ); | 214 | floatx80 floatx80_rem( struct roundingData *, floatx80, floatx80 ); |
| 225 | floatx80 floatx80_sqrt( floatx80 ); | 215 | floatx80 floatx80_sqrt( struct roundingData *, floatx80 ); |
| 226 | char floatx80_eq( floatx80, floatx80 ); | 216 | char floatx80_eq( floatx80, floatx80 ); |
| 227 | char floatx80_le( floatx80, floatx80 ); | 217 | char floatx80_le( floatx80, floatx80 ); |
| 228 | char floatx80_lt( floatx80, floatx80 ); | 218 | char floatx80_lt( floatx80, floatx80 ); |
diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c index b801cd66b6ea..9b367a65cb4d 100644 --- a/arch/arm/vfp/vfpdouble.c +++ b/arch/arm/vfp/vfpdouble.c | |||
| @@ -770,6 +770,9 @@ vfp_double_add(struct vfp_double *vdd, struct vfp_double *vdn, | |||
| 770 | if ((s64)m_sig < 0) { | 770 | if ((s64)m_sig < 0) { |
| 771 | vdd->sign = vfp_sign_negate(vdd->sign); | 771 | vdd->sign = vfp_sign_negate(vdd->sign); |
| 772 | m_sig = -m_sig; | 772 | m_sig = -m_sig; |
| 773 | } else if (m_sig == 0) { | ||
| 774 | vdd->sign = (fpscr & FPSCR_RMODE_MASK) == | ||
| 775 | FPSCR_ROUND_MINUSINF ? 0x8000 : 0; | ||
| 773 | } | 776 | } |
| 774 | } else { | 777 | } else { |
| 775 | m_sig += vdn->significand; | 778 | m_sig += vdn->significand; |
diff --git a/drivers/char/watchdog/sa1100_wdt.c b/drivers/char/watchdog/sa1100_wdt.c index 1b2132617dc3..fb88b4041dca 100644 --- a/drivers/char/watchdog/sa1100_wdt.c +++ b/drivers/char/watchdog/sa1100_wdt.c | |||
| @@ -36,13 +36,10 @@ | |||
| 36 | #include <asm/uaccess.h> | 36 | #include <asm/uaccess.h> |
| 37 | 37 | ||
| 38 | #define OSCR_FREQ CLOCK_TICK_RATE | 38 | #define OSCR_FREQ CLOCK_TICK_RATE |
| 39 | #define SA1100_CLOSE_MAGIC (0x5afc4453) | ||
| 40 | 39 | ||
| 41 | static unsigned long sa1100wdt_users; | 40 | static unsigned long sa1100wdt_users; |
| 42 | static int expect_close; | ||
| 43 | static int pre_margin; | 41 | static int pre_margin; |
| 44 | static int boot_status; | 42 | static int boot_status; |
| 45 | static int nowayout = WATCHDOG_NOWAYOUT; | ||
| 46 | 43 | ||
| 47 | /* | 44 | /* |
| 48 | * Allow only one person to hold it open | 45 | * Allow only one person to hold it open |
| @@ -62,55 +59,33 @@ static int sa1100dog_open(struct inode *inode, struct file *file) | |||
| 62 | } | 59 | } |
| 63 | 60 | ||
| 64 | /* | 61 | /* |
| 65 | * Shut off the timer. | 62 | * The watchdog cannot be disabled. |
| 66 | * Lock it in if it's a module and we defined ...NOWAYOUT | 63 | * |
| 67 | * Oddly, the watchdog can only be enabled, but we can turn off | 64 | * Previous comments suggested that turning off the interrupt by |
| 68 | * the interrupt, which appears to prevent the watchdog timing out. | 65 | * clearing OIER[E3] would prevent the watchdog timing out but this |
| 66 | * does not appear to be true (at least on the PXA255). | ||
| 69 | */ | 67 | */ |
| 70 | static int sa1100dog_release(struct inode *inode, struct file *file) | 68 | static int sa1100dog_release(struct inode *inode, struct file *file) |
| 71 | { | 69 | { |
| 72 | OSMR3 = OSCR + pre_margin; | 70 | printk(KERN_CRIT "WATCHDOG: Device closed - timer will not stop\n"); |
| 73 | |||
| 74 | if (expect_close == SA1100_CLOSE_MAGIC) { | ||
| 75 | OIER &= ~OIER_E3; | ||
| 76 | } else { | ||
| 77 | printk(KERN_CRIT "WATCHDOG: WDT device closed unexpectedly. WDT will not stop!\n"); | ||
| 78 | } | ||
| 79 | 71 | ||
| 80 | clear_bit(1, &sa1100wdt_users); | 72 | clear_bit(1, &sa1100wdt_users); |
| 81 | expect_close = 0; | ||
| 82 | 73 | ||
| 83 | return 0; | 74 | return 0; |
| 84 | } | 75 | } |
| 85 | 76 | ||
| 86 | static ssize_t sa1100dog_write(struct file *file, const char *data, size_t len, loff_t *ppos) | 77 | static ssize_t sa1100dog_write(struct file *file, const char *data, size_t len, loff_t *ppos) |
| 87 | { | 78 | { |
| 88 | if (len) { | 79 | if (len) |
| 89 | if (!nowayout) { | ||
| 90 | size_t i; | ||
| 91 | |||
| 92 | expect_close = 0; | ||
| 93 | |||
| 94 | for (i = 0; i != len; i++) { | ||
| 95 | char c; | ||
| 96 | |||
| 97 | if (get_user(c, data + i)) | ||
| 98 | return -EFAULT; | ||
| 99 | if (c == 'V') | ||
| 100 | expect_close = SA1100_CLOSE_MAGIC; | ||
| 101 | } | ||
| 102 | } | ||
| 103 | /* Refresh OSMR3 timer. */ | 80 | /* Refresh OSMR3 timer. */ |
| 104 | OSMR3 = OSCR + pre_margin; | 81 | OSMR3 = OSCR + pre_margin; |
| 105 | } | ||
| 106 | 82 | ||
| 107 | return len; | 83 | return len; |
| 108 | } | 84 | } |
| 109 | 85 | ||
| 110 | static struct watchdog_info ident = { | 86 | static struct watchdog_info ident = { |
| 111 | .options = WDIOF_CARDRESET | WDIOF_MAGICCLOSE | | 87 | .options = WDIOF_CARDRESET | WDIOF_SETTIMEOUT | WDIOF_KEEPALIVEPING, |
| 112 | WDIOF_SETTIMEOUT | WDIOF_KEEPALIVEPING, | 88 | .identity = "SA1100/PXA255 Watchdog", |
| 113 | .identity = "SA1100 Watchdog", | ||
| 114 | }; | 89 | }; |
| 115 | 90 | ||
| 116 | static int sa1100dog_ioctl(struct inode *inode, struct file *file, | 91 | static int sa1100dog_ioctl(struct inode *inode, struct file *file, |
| @@ -172,7 +147,7 @@ static struct file_operations sa1100dog_fops = | |||
| 172 | static struct miscdevice sa1100dog_miscdev = | 147 | static struct miscdevice sa1100dog_miscdev = |
| 173 | { | 148 | { |
| 174 | .minor = WATCHDOG_MINOR, | 149 | .minor = WATCHDOG_MINOR, |
| 175 | .name = "SA1100/PXA2xx watchdog", | 150 | .name = "watchdog", |
| 176 | .fops = &sa1100dog_fops, | 151 | .fops = &sa1100dog_fops, |
| 177 | }; | 152 | }; |
| 178 | 153 | ||
| @@ -194,7 +169,6 @@ static int __init sa1100dog_init(void) | |||
| 194 | if (ret == 0) | 169 | if (ret == 0) |
| 195 | printk("SA1100/PXA2xx Watchdog Timer: timer margin %d sec\n", | 170 | printk("SA1100/PXA2xx Watchdog Timer: timer margin %d sec\n", |
| 196 | margin); | 171 | margin); |
| 197 | |||
| 198 | return ret; | 172 | return ret; |
| 199 | } | 173 | } |
| 200 | 174 | ||
| @@ -212,8 +186,5 @@ MODULE_DESCRIPTION("SA1100/PXA2xx Watchdog"); | |||
| 212 | module_param(margin, int, 0); | 186 | module_param(margin, int, 0); |
| 213 | MODULE_PARM_DESC(margin, "Watchdog margin in seconds (default 60s)"); | 187 | MODULE_PARM_DESC(margin, "Watchdog margin in seconds (default 60s)"); |
| 214 | 188 | ||
| 215 | module_param(nowayout, int, 0); | ||
| 216 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started"); | ||
| 217 | |||
| 218 | MODULE_LICENSE("GPL"); | 189 | MODULE_LICENSE("GPL"); |
| 219 | MODULE_ALIAS_MISCDEV(WATCHDOG_MINOR); | 190 | MODULE_ALIAS_MISCDEV(WATCHDOG_MINOR); |
