TC1784
CPU Subsystem
User´s Manual
2-85
V1.1, 2011-05
CPU, V3.03
load-store interface will generate unaligned accesses (16-bit aligned), which will result
in up to 64-bits of data being transferred to or from the CPU (for non-context operations).
If the data access is made within a DCache line, no matter the alignment, and a cache
hit is detected then the requested data is returned to the CPU in a single cycle. If the data
access is made to the end of a DCache line, such that the requested data would span
two DCache lines, a single wait cycle is incurred (if both cache lines are present in the
cache, otherwise a refill sequence is required for the missing cache line(s)).
The TC1784 data cache is of the writeback type. When the CPU writes to a cacheable
location the data is merged with the corresponding cache line and not written to main
memory immediately. Associated with each cache line is a single ‘dirty’ bit, to denote that
the data in the cache line has been modified. Whenever a CPU load-store access results
in a cache miss, and each of the potential cache ways that could hold the requested
cache line are valid, one of the cache lines is chosen for eviction based upon the LRU
replacement algorithm. The line selected for eviction is then checked to determine if it
has been modified using its dirty bit. If the line has not been modified the line is discarded
and the refill sequence started immediately. If the line has been modified then the dirty
data is first written back to main memory before the refill is initiated. Due to the single
dirty bit per cache line, 128 bits of data will always be written back, resulting in LMB Burst
Transfer 2 (BTR2) transactions.
Data Cache refills always result in the full cache line being refilled, with the critical
double-word of the DCache line being fetched first. A refill sequence will always affect
only one cache line. There is no prefetching of the next cache line. Due to the uniform
size of DCache refill sequences, such refills are always implemented using LMB Burst
Transfer 2 (BTR2) transactions.
2.15.5
Data Line Buffer
The DMI module contains a 128-bit Data Line Buffer (DLB).
When the TC1784 device is configured to contain a data cache, the DLB acts as a
streaming buffer for the main data cache. Data cache refill sequences transfer data to
the DLB and then immediately to the TriCore CPU without updating the main data cache.
The DLB contents are then transferred to the main data cache on the next data cache
miss. The presence of the DLB is transparent to software but ensures that accesses to
cacheable addresses typically
1)
incur no more wait cycles than accesses to non-
cacheable addresses.
When the TC1784 device is configured without a data cache, the DLB acts as a single
line data cache for accesses to cacheable addresses.
1) No additional wait states are incurred as long as the data cache line to be replaced does not contain dirty
information which requires a cache line writeback.
Содержание TC1784
Страница 1: ...User s Manual V1 1 2011 05 Microcontrollers TC1784 32 Bit Single Chip Microcontroller ...
Страница 3: ...User s Manual V1 1 2011 05 Microcontrollers TC1784 32 Bit Single Chip Microcontroller ...
Страница 950: ...TC1784 Direct Memory Access Controller DMA User s Manual 11 132 V1 1 2011 05 DMA V3 03 ...
Страница 1949: ...TC1784 General Purpose Timer Array GPTA v5 User s Manual 21 297 V1 1 2011 05 GPTA v5 V1 14 ...
Страница 2350: ...w w w i n f i n e o n c o m Published by Infineon Technologies AG Doc_Number ...