summaryrefslogtreecommitdiffstats
path: root/mpi/mips3/README
diff options
context:
space:
mode:
Diffstat (limited to 'mpi/mips3/README')
-rw-r--r--mpi/mips3/README23
1 files changed, 23 insertions, 0 deletions
diff --git a/mpi/mips3/README b/mpi/mips3/README
new file mode 100644
index 000000000..e94b2c746
--- /dev/null
+++ b/mpi/mips3/README
@@ -0,0 +1,23 @@
+This directory contains mpn functions optimized for MIPS3. Example of
+processors that implement MIPS3 are R4000, R4400, R4600, R4700, and R8000.
+
+RELEVANT OPTIMIZATION ISSUES
+
+1. On the R4000 and R4400, branches, both the plain and the "likely" ones,
+ take 3 cycles to execute. (The fastest possible loop will take 4 cycles,
+ because of the delay insn.)
+
+ On the R4600, branches takes a single cycle
+
+ On the R8000, branches often take no noticable cycles, as they are
+ executed in a separate function unit..
+
+2. The R4000 and R4400 have a load latency of 4 cycles.
+
+3. On the R4000 and R4400, multiplies take a data-dependent number of
+ cycles, contrary to the SGI documentation. There seem to be 3 or 4
+ possible latencies.
+
+STATUS
+
+Good...