<div dir="ltr">I'm sorry, the text I was thinking of was on page 20 of the Iterated Register Coalescing paper (<a href="http://www.cs.cmu.edu/afs/cs/academic/class/15745-s07/www/papers/george.pdf">http://www.cs.cmu.edu/afs/cs/academic/class/15745-s07/www/papers/george.pdf</a>), "Move instructions are given special consideration...", but it's still the same idea.</div> <div class="gmail_extra"> <div class="gmail_quote">On Fri, Dec 27, 2013 at 7:38 PM, Connor Abbott <<a href="mailto:cwabbott0@gmail.com" target="_blank">cwabbott0@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I'm not sure if this is relevant to the i965 fs backend in particular, but Briggs in his thesis [1] recommended simply ignoring the liveness of any register read by a copy instruction immediately after said instruction when computing the interference graph in order to solve this exact problem (see section 2.2.2). [1] <a href="http://www.cs.utexas.edu/~mckinley/380C/lecs/briggs-thesis-1992.pdf" target="_blank">http://www.cs.utexas.edu/~mckinley/380C/lecs/briggs-thesis-1992.pdf</a><div class="HOEnZb"><div class="h5"> <div class="gmail_quote">On Dec 19, 2013 4:40 PM, "Matt Turner" <<a href="mailto:mattst88@gmail.com" target="_blank">mattst88@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Previously we simply considered two registers whose live ranges overlapped to interfere. Cases such as set A ------ ... | mov B, A -- | ... | B | A use B -- | ... | use A ------ would be considered to interfere, even though B is an unmodified copy of A whose live range fit wholly inside that of A. If no writes to A or B occur between the mov B, A and the use of B then we can safely coalesce them. Instead of removing MOV instructions, we make them NOPs and remove them at once after the main pass is finished in order to avoid recomputing live intervals (which are needed to perform the previous step). total instructions in shared programs: 1543768 -> 1513077 (-1.99%) instructions in affected programs: 951563 -> 920872 (-3.23%) GAINED: 46 LOST: 22 --- src/mesa/drivers/dri/i965/brw_fs.cpp | 69 ++++++++++++++++++++++++++++++++---- 1 file changed, 62 insertions(+), 7 deletions(-) diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp index e4ac0a5..ad56b87 100644 --- a/src/mesa/drivers/dri/i965/brw_fs.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp @@ -2273,7 +2273,7 @@ fs_visitor::register_coalesce() int last_use[MAX_SAMPLER_MESSAGE_SIZE]; int next_ip = 0; - foreach_list_safe(node, &this->instructions) { + foreach_list(node, &this->instructions) { fs_inst *inst = (fs_inst *)node; int ip = next_ip; @@ -2299,8 +2299,39 @@ fs_visitor::register_coalesce() int var_to = live_intervals->var_from_reg(&inst->dst); if (live_intervals->vars_interfere(var_from, var_to) && - !inst->dst.equals(inst->src[0])) - continue; + !inst->dst.equals(inst->src[0])) { + + if (live_intervals->end[var_to] > live_intervals->end[var_from]) + continue; + + bool overwritten = false; + int scan_ip = -1; + + foreach_list(n, &this->instructions) { + fs_inst *scan_inst = (fs_inst *)n; + scan_ip++; + + if (scan_inst->is_control_flow()) { + overwritten = true; + break; + } + + if (scan_ip <= live_intervals->start[var_to]) + continue; + + if (scan_ip > live_intervals->end[var_to]) + break; + + if (scan_inst->dst.equals(inst->dst) || + scan_inst->dst.equals(inst->src[0])) { + overwritten = true; + break; + } + } + + if (overwritten) + continue; + } if (reg_from != inst->src[0].reg) { reg_from = inst->src[0].reg; @@ -2342,9 +2373,18 @@ fs_visitor::register_coalesce() if (live_channels_remaining) continue; + bool removed = false; for (int i = 0; i < src_size; i++) { - if (mov[i]) - mov[i]->remove(); + if (mov[i]) { + removed = true; + + mov[i]->opcode = BRW_OPCODE_NOP; + mov[i]->conditional_mod = BRW_CONDITIONAL_NONE; + mov[i]->dst = reg_undef; + mov[i]->src[0] = reg_undef; + mov[i]->src[1] = reg_undef; + mov[i]->src[2] = reg_undef; + } } foreach_list(node, &this->instructions) { @@ -2366,11 +2406,26 @@ fs_visitor::register_coalesce() scan_inst->src[j].reg_offset = reg_to_offset[i]; } } - - progress = true; } } } + + if (removed) { + live_intervals->start[var_to] = MIN2(live_intervals->start[var_to], + live_intervals->start[var_from]); + live_intervals->end[var_to] = MAX2(live_intervals->end[var_to], + live_intervals->end[var_from]); + reg_from = -1; + } + } + + foreach_list_safe(node, &this->instructions) { + fs_inst *inst = (fs_inst *)node; + + if (inst->opcode == BRW_OPCODE_NOP) { + inst->remove(); + progress = true; + } } if (progress) -- 1.8.3.2 _______________________________________________ mesa-dev mailing list <a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank">mesa-dev@lists.freedesktop.org</a> <a href="http://lists.freedesktop.org/mailman/listinfo/mesa-dev" target="_blank">http://lists.freedesktop.org/mailman/listinfo/mesa-dev</a> </blockquote></div> </div></div></blockquote></div> </div>