Add benchmarks using iai-callgrind

This crate [1] makes it reasonably easy to get instruction count
performance metrics that are stable enough to run in CI, and has worked
out well since integrating it with `libm`. Add new benchmarks for `mem`
functions using `iai-callgrind`, modeling them off of the existing
benchmarks.

[1]: https://github.com/iai-callgrind/iai-callgrind

Author: tgross35

testcrate/Cargo.toml

@@ -16,6 +16,8 @@ doctest = false
 rand_xoshiro = "0.6"
 # To compare float builtins against
 rustc_apfloat = "0.2.1"
+# Really a dev dependency, but dev dependencies can't be optional
+iai-callgrind = { version = "0.14.0", optional = true }
 
 [dependencies.compiler_builtins]
 path = "../compiler-builtins"
@@ -47,9 +49,16 @@ no-sys-f16-f64-convert = []
 # Skip tests that rely on f16 symbols being available on the system
 no-sys-f16 = ["no-sys-f16-f64-convert"]
 
+# Enable icount benchmarks (requires iai-callgrind and valgrind)
+icount = ["dep:iai-callgrind"]
+
 # Enable report generation without bringing in more dependencies by default
 benchmarking-reports = ["criterion/plotters", "criterion/html_reports"]
 
+# NOTE: benchmarks must be run with `--no-default-features` or with
+# `-p testcrate`, otherwise the default `compiler-builtins` feature of the
+# `compiler_builtins` crate gets activated, resulting in linker errors.
+
 [[bench]]
 name = "float_add"
 harness = false
@@ -85,3 +94,8 @@ harness = false
 [[bench]]
 name = "float_pow"
 harness = false
+
+[[bench]]
+name = "mem_icount"
+harness = false
+required-features = ["icount"]