floatx.asm: add specific tests for exponent boundary conditions

Add tests for the exponent boundaries (underflow, denorm, normal).
This commit is contained in:
H. Peter Anvin 2007-10-04 23:09:19 -07:00
parent cdb227fa71
commit 6b8abf54d3

View File

@ -23,6 +23,11 @@
dw 0x0.0000123456789
dw 0x0.0000123456789p+10
dw 0x0.0000123456789p-10
dw 0x1.0p-25 ; Underflow
dw 0x1.0p-24 ; Smallest denorm
dw 0x1.0p-15 ; Denorm
dw 0x1.0p-14 ; Smallest non-denorm
dw 0x1.0p+15 ; Biggest possible exponent
; 32-bit
dd 1.0
@ -48,6 +53,11 @@
dd 0x0000123456789.0
dd 0x123456789.0p+0
dd 0x123456789.0p+64
dd 0x1.0p-150 ; Underflow
dd 0x1.0p-149 ; Smallest denorm
dd 0x1.0p-127 ; Denorm
dd 0x1.0p-126 ; Smallest non-denorm
dd 0x1.0p+127 ; Biggest possible exponent
; 64-bit
dq 1.0
@ -73,6 +83,11 @@
dq 0x0000123456789.0
dq 0x123456789.0p+0
dq 0x123456789.0p+300
dq 0x1.0p-1075 ; Underflow
dq 0x1.0p-1074 ; Smallest denorm
dq 0x1.0p-1023 ; Denorm
dq 0x1.0p-1022 ; Smallest non-denorm
dq 0x1.0p+1023 ; Biggest possible exponent
; 80-bit
dt 1.0
@ -98,6 +113,12 @@
dt 0x0000123456789.0
dt 0x123456789.0p+0
dt 0x123456789.0p+1024
dt 0x1.0p-16446 ; Underflow
dt 0x1.0p-16445 ; Smallest denorm
dt 0x1.0p-16383 ; Denorm
dt 0x1.0p-16382 ; Smallest non-denorm
dt 0x1.0p+16383 ; Biggest possible exponent
; 128-bit
do 1.0
@ -123,3 +144,8 @@
do 0x0000123456789.0
do 0x123456789.0p+0
do 0x123456789.0p+1024
do 0x1.0p-16495 ; Underflow
do 0x1.0p-16494 ; Smallest denorm
do 0x1.0p-16383 ; Denorm
do 0x1.0p-16382 ; Smallest non-denorm
do 0x1.0p+16383 ; Biggest possible exponent