diff options
| -rw-r--r-- | generic/tclExecute.c | 37 |
1 files changed, 16 insertions, 21 deletions
diff --git a/generic/tclExecute.c b/generic/tclExecute.c index ae9d0c7..bc755e8 100644 --- a/generic/tclExecute.c +++ b/generic/tclExecute.c @@ -2300,16 +2300,10 @@ TEBCresume( TCL_DTRACE_INST_NEXT(); - /* - * These two instructions account for 26% of all instructions (according - * to measurements on tclbench by Ben Vitale - * [http://www.cs.toronto.edu/syslab/pubs/tcl2005-vitale-zaleski.pdf] - * Resolving them before the switch reduces the cost of branch - * mispredictions, seems to improve runtime by 5% to 15%, and (amazingly!) - * reduces total obj size. - */ - - if (*pc == INST_START_CMD) { + while (*pc == INST_START_CMD) { +#ifdef TCL_COMPILE_STATS + iPtr->stats.instructionCount[*pc]++; +#endif iPtr->cmdCount += TclGetUInt4AtPtr(pc+5); if (checkInterp) { checkInterp = 0; @@ -2321,6 +2315,15 @@ TEBCresume( pc += 9; } + /* + * These two instructions account for 26% of all instructions (according + * to measurements on tclbench by Ben Vitale + * [http://www.cs.toronto.edu/syslab/pubs/tcl2005-vitale-zaleski.pdf] + * Resolving them before the switch reduces the cost of branch + * mispredictions, seems to improve runtime by 5% to 15%, and (amazingly!) + * reduces total obj size. + */ + if (*pc == INST_LOAD_SCALAR1) { goto instLoadScalar1; } else if (*pc == INST_PUSH1) { @@ -2503,19 +2506,10 @@ TEBCresume( TRACE_WITH_OBJ(("=> discarding "), OBJ_AT_TOS); objPtr = POP_OBJECT(); TclDecrRefCount(objPtr); - - /* - * Runtime peephole optimisation: an INST_POP is scheduled at the end - * of most commands. If the next instruction is an INST_START_CMD, - * fall through to it. - */ - - pc++; - NEXT_INST_F(0, 0, 0); + NEXT_INST_F(1, 0, 0); case INST_NOP: - pc += 1; - goto cleanup0; + NEXT_INST_F(1, 0, 0); case INST_DUP: objResultPtr = OBJ_AT_TOS; @@ -7081,6 +7075,7 @@ TEBCresume( { const char *bytes; + checkInterp = 1; length = 0; /* |