Add benchmarks for floating point math

This adds comparisons among the compiler-builtins function, system
functions if available, and optionally handwritten assembly.

These also help us identify inconsistencies between this crate and
system functions, which may otherwise go unnoticed if intrinsics get
lowered to inline operations rather than library calls.

Author: tgross35

ci/run.sh

@@ -4,7 +4,9 @@ set -eux
 
 target="${1:-}"
 
-if [ -z "${1:-}" ]; then
+export RUST_BACKTRACE="${RUST_BACKTRACE:-full}"
+
+if [ -z "$target" ]; then
     host_target=$(rustc -vV | awk '/^host/ { print $2 }')
     echo "Defaulted to host target $host_target"
     target="$host_target"
@@ -30,6 +32,8 @@ else
     $run --features no-asm --release
     $run --features no-f16-f128
     $run --features no-f16-f128 --release
+    $run --benches
+    $run --benches --release
 fi
 
 if [ "${TEST_VERBATIM:-}" = "1" ]; then

testcrate/Cargo.toml

@@ -21,6 +21,10 @@ path = ".."
 default-features = false
 features = ["public-test-deps"]
 
+[dev-dependencies]
+criterion = { version = "0.5.1", default-features = false, features = ["cargo_bench_support"] }
+paste = "1.0.15"
+
 [target.'cfg(all(target_arch = "arm", not(any(target_env = "gnu", target_env = "musl")), target_os = "linux"))'.dev-dependencies]
 test = { git = "https://github.com/japaric/utest" }
 utest-cortex-m-qemu = { default-features = false, git = "https://github.com/japaric/utest" }
@@ -34,6 +38,48 @@ no-f16-f128 = ["compiler_builtins/no-f16-f128"]
 mem = ["compiler_builtins/mem"]
 mangled-names = ["compiler_builtins/mangled-names"]
 # Skip tests that rely on f128 symbols being available on the system
-no-sys-f128 = ["no-sys-f128-int-convert"]
+no-sys-f128 = ["no-sys-f128-int-convert", "no-sys-f16-f128-convert"]
 # Some platforms have some f128 functions but everything except integer conversions
 no-sys-f128-int-convert = []
+no-sys-f16-f128-convert = []
+# Skip tests that rely on f16 symbols being available on the system
+no-sys-f16 = []
+
+# Enable report generation without bringing in more dependencies by default
+benchmarking-reports = ["criterion/plotters", "criterion/html_reports"]
+
+[[bench]]
+name = "float_add"
+harness = false
+
+[[bench]]
+name = "float_sub"
+harness = false
+
+[[bench]]
+name = "float_mul"
+harness = false
+
+[[bench]]
+name = "float_div"
+harness = false
+
+[[bench]]
+name = "float_cmp"
+harness = false
+
+[[bench]]
+name = "float_conv"
+harness = false
+
+[[bench]]
+name = "float_extend"
+harness = false
+
+[[bench]]
+name = "float_trunc"
+harness = false
+
+[[bench]]
+name = "float_pow"
+harness = false