LLVM propeller and Node.js with 2 workloads

gistfile1.txt

Compiler	LLVM v10.0	(branch: plo-dev, commit id: 2031626e55e6f8cd67aafb53142ada64a97db0d1)

Workload 1: webtooling (https://github.com/v8/web-tooling-benchmark)
Webtooling is simple to run. Only thing you need is to build node binary (Node.js runtime) 

	                               node-orignal	 node-with-propeller	with-propeller/orig  
score (runs/s)	                        10.15	            9.53	      0.939 <---- 7% regression in absolute throughput
binary-size (Bytes)	               66,552,488   	  75,208,088 	      1.130 <--- size increased
cycles	                      403,260,331,520	 419,827,132,716	      1.041 <--- 4% more cycles
instructions	                701,961,055,882	 700,658,118,964	      0.998 <--- instructions fairely remained same
icache_64b.iftag_stall	       16,069,917,504	   9,930,932,775	      0.618 <--- 40% improvement
itlb_misses.walk_completed	       55,644,425	      55,759,159	      1.002 
itlb_misses.walk_completed_4k	     55,302,379	      55,366,957	      1.001
br_misp_retired.all_branches	  2,089,716,437	   2,081,746,644	      0.996
br_inst_retired.all_branches	151,219,359,616	 168,375,100,473	      1.113 <-- 11% branches retired (regression)
			
metric_ITLB_Misses (%)	                 3.98	            2.37	      0.595 <--- front-end bottleneck due to iTLB miss reduction
metric_ITLB_MPKI(%)	                    0.079	           0.079	      1.000
metric_ITLB_4K_MPKI(%)	                0.079	           0.079	      1.000
metric_branch mispredict ratio	         1.38	            1.24	      0.896


Workload 2: Ghost.js ( https://github.com/TryGhost/Ghost.git)
Ghost.js involves 3 components (node application, database and apache client "ab")

	                               node-orignal	 node-with-propeller	wp/orig
Elapsed Time	                             50	              50	      1.000
RPS	                                   124.34            114.3	      0.919  <--- 9% regression
binary-size (Bytes)	               66,552,488       63,593,208 	      0.956  <--- improvement (after running split on the binary)
cycles	                      186,369,907,075	 188,624,952,924	      1.012  <--- 1% code increase
instructions	                299,760,152,781	 282,230,632,344	      0.942  <--- 6% improvement
icache_64b.iftag_stall	       12,336,211,838	  11,179,865,194	      0.906  <--- 10% improvement
itlb_misses.walk_completed	       65,090,576	      54,891,314	      0.843  <--- 16% improvement
itlb_misses.walk_completed_4k	     64,916,461	      54,869,388	      0.845  <---
br_misp_retired.all_branches	  1,112,011,928	   1,059,303,532	      0.953  <--- 5% improvement
br_inst_retired.all_branches	 63,050,053,963	  67,215,061,774	      1.066  <--- 6% more branches (regression)
			
metric_ITLB_Misses (%)	                 6.62	            5.93	      0.896  <--- 11% improvement
metric_ITLB_MPKI(%)	                    0.217	           0.194	      0.894
metric_ITLB_4K_MPKI(%)	               0.2165	           0.194	      0.898
metric_branch mispredict ratio	         1.76	            1.58	      0.898