Current benchmark results, Release shell, test case as above but tweaked a litte (will be updated), fib(35): On x64 Xeon: Ion: 70ms Baseline: 100ms Baseline with old debug code, debugging not enabled: 155ms Baseline with old debug code, debugging enabled on unrelated function: 155ms Baseline with new debug code, debugging not enabled: 245ms (1.6x) Baseline with new debug code, debugging enabled on unrelated function: 655ms (2.7x slowdown over new no-debug case, 4.2x slowdown over old) On ARM64 M1: Ion: 66ms Baseline: 78ms Baseline with old debug code, debugging not enabled: 133ms Baseline with old debug code, debugging enabled on unrelated function: 133ms Baseline with new debug code, debugging not enabled: 196ms (1.5x) Baseline with new debug code, debugging enabled on unrelated function: 330ms (1.7x over new no-debug case, 2.5x over old) Once debugging is enabled for a function we'll make a call to C++ filtering for every site in the function, so setting a breakpoint in a function will slow that function down greatly (not measured, but earlier results suggest an order of magnitude slowdown).
Bug 1756951 Comment 12 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
Current benchmark results, Release shell, test case as above but tweaked a litte (will be updated), fib(35): On x64 Xeon: Ion: 70ms Baseline: 100ms Baseline with old debug code, debugging not enabled: 155ms Baseline with old debug code, debugging enabled on unrelated function: 155ms Baseline with new debug code, debugging not enabled: 245ms (1.6x) Baseline with new debug code, debugging enabled on unrelated function: 655ms (2.7x slowdown over new no-debug case, 4.2x slowdown over old) On ARM64 M1: Ion: 66ms Baseline: 78ms Baseline with old debug code, debugging not enabled: 133ms Baseline with old debug code, debugging enabled on unrelated function: 133ms Baseline with new debug code, debugging not enabled: 196ms (1.5x) Baseline with new debug code, debugging enabled on unrelated function: 330ms (1.7x over new no-debug case, 2.5x over old) Once debugging is enabled for a function we'll make a call to C++ filtering for every site in the function, so setting a breakpoint in a function will slow that function down greatly (not measured, but earlier results suggest three orders of magnitude slowdown).