English
Language : 

SMJ34020A_06 Datasheet, PDF (10/97 Pages) Texas Instruments – GRAPHICS SYSTEM PROCESSOR
SMJ34020A
GRAPHICS SYSTEM PROCESSOR
SGUS011D − APRIL 1991 − REVISED SEPTEMBER 2004
instruction cache
An on-chip cache contains 512 bytes of RAM and provides unimpeded access to instructions. The cache
operates automatically and is transparent to software. The cache is divided into four 128-byte segments.
Associated with each segment is a 22-bit segment start address register (SSA) to identify the addresses in
memory corresponding to the current contents of the cache segment. Each cache segment is further partitioned
into eight subsegments of four long words (32 bits) each. Each subsegment has an associated present (P) flag
to indicate whether or not the subsegment contains valid data.
The cache is loaded only when an instruction requested by the execution section of the SMJ34020A is not
already contained within the cache. A least-recently-used (LRU) algorithm determines which of the four
segments of the cache is overwritten with new data. For this purpose, an internal four-by-two LRU stack keeps
track of cache usage. Although the cache is loaded so as to always fill a subsegment completely, not all eight
subsegments within a segment are necessarily filled (this is dependent upon the instruction stream).
status register
The status register (ST) is a special purpose 32-bit register dedicated to status codes set by the results of implicit
and explicit compare operations and parameters used to specify the length and behavior of fields 0 and 1. During
an interrupt, when the IX bit in the ST is placed on the stack, it indicates that execution of an interruptable
instruction (PIXBLT, FILL or LINE) was halted to service the interrupt. The single-step bit causes a trap to the
single-step vector (located at address FFFF FBE0h) after the execution of one instruction when the bit is set
high. Normal program execution occurs when the bit is set low.
fields, bytes, words, long words, pixels and pixel arrays
The SMJ34020A outputs a 28-bit address on LAD4−LAD31 that is valid at the falling edge of ALTCH. The most
significant 27 bits (LAD5 −LAD31) define a 32-bit-long word of physical memory; logically, however, the
SMJ34020A views memory data as fields addressable at the bit level. The least significant bit of the 28-bit
address (LAD4) is used to select the odd or even word when accessing 16-bit memories (indicated by SIZE16
asserted low). Primitive data types supported by the SMJ34020A include bytes, words, long words, pixels, two
independent fields of from 1 to 32 bits, and user-defined pixel arrays.
Words and long words, respectively, refer to 16- and 32-bit values that are aligned on 32-bit boundaries.
The two independent fields are referenced as field 0 and field 1. The attributes of these fields (field size and sign
extension within a register) are defined in the status register as FS0, FE0, FS1, and FE1. Fields 0 and 1 are
specified independently to be signed or unsigned and from 1 to 32 bits in length. Bytes are special 8-bit cases
of the field data type, while pixels are 1, 2, 4, 8, 16, or 32 bits in length. In general, fields (including bytes) can
start and terminate on arbitrary bit boundaries; however, pixels must pack evenly into 32-bit-long words.
pixel operations
Pixel arrays are two-dimensional data types of user-defined width, length, pixel depth (number of bits per pixel),
and pitch (distance between rows). A pixel or pixel array can be accessed by means of either its memory address
or its XY coordinates. Transfers of individual pixels or pixel blocks are influenced by the pixel processing,
transparency, window checking, plane masking, pixel masking, or corner adjustment operations selected. For
further information, see the TMS32020 User’s Guide, literature number SPVU019.
10
• POST OFFICE BOX 1443 HOUSTON, TEXAS 77251−1443