counted-float 0.9.4__tar.gz → 0.9.6__tar.gz

counted_float-0.9.6/PKG-INFO

@@ -0,0 +1,629 @@
+Metadata-Version: 2.4
+Name: counted-float
+Version: 0.9.6
+Summary: Count floating-point operations in Python code & benchmark relative flop costs.
+Project-URL: Source, https://github.com/bertpl/counted-float
+Project-URL: ChangeLog, https://github.com/bertpl/counted-float/blob/develop/CHANGELOG.md
+Project-URL: Issues, https://github.com/bertpl/counted-float/issues
+Project-URL: Roadmap, https://github.com/bertpl/counted-float/milestones
+License-File: LICENSE
+Requires-Python: >=3.11
+Requires-Dist: numpy>=1.20
+Requires-Dist: psutil>=5.0
+Requires-Dist: py-cpuinfo>=9.0.0
+Requires-Dist: pydantic>=2.0.0
+Requires-Dist: rich>=13.0.0
+Provides-Extra: cli
+Requires-Dist: click>=8.0.0; extra == 'cli'
+Provides-Extra: numba
+Requires-Dist: numba>=0.50; extra == 'numba'
+Description-Content-Type: text/markdown
+
+<!--START_SECTION:images-->
+![shields.io-python-versions](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13-blue)
+![genbadge-test-count](https://bertpl.github.io/counted-float/version_artifacts/v0.9.6/badge-test-count.svg)
+![genbadge-test-coverage](https://bertpl.github.io/counted-float/version_artifacts/v0.9.6/badge-coverage.svg)
+![counted_float logo](https://bertpl.github.io/counted-float/version_artifacts/v0.9.6/splash.webp)
+<!--END_SECTION:images-->
+
+# counted-float
+
+This Python package provides functionality for...
+- **counting floating point operations** (FLOPs) of numerical algorithms implemented in plain Python, optionally weighted by their relative cost of execution
+- **running benchmarks** to estimate the relative cost of executing various floating-point operations (requires `numba` optional dependency for achieving accurate results)
+
+The target application area is evaluation of research prototypes of numerical algorithms where (weighted) flop counting can be 
+useful for estimating total computational cost, in cases where benchmarking a compiled version (C, Rust, ...) is not 
+feasible or desirable.
+
+Flop weights are computed using a highly curated dataset spanning a wide range of modern CPUs:
+- 19 benchmarks, 16 spec sheets, 12 third party measurements (Agner Fog, uops.info)
+- covering x86 (Intel, AMD) and ARM (Apple, AWS, Azure) architectures
+
+# 1. Installation
+
+Use you favorite package manager such as `uv` or `pip`:
+
+```
+pip install counted-float           # install without numba optional dependency
+pip install counted-float[numba]    # install with numba optional dependency
+```
+Numba is optional due to its relatively large size (40-50MB, including llvmlite), but without it, benchmarks will
+not be reliable (but will still run, but not in jit-compiled form).
+
+NOTE: the `cli` optional dependency is only useful when installing the code as a tool using e.g. `uv` or `pipx` (see below)
+
+# 2. Counting Flops
+
+## 2.1. CountedFloat class
+
+In order to instrument all floating point operations with counting functionality,
+the `CountedFloat` class was implemented, which is a drop-in replacement for the built-in `float` type.
+The `CountedFloat` class is a subclass of `float` and is "contagious", meaning that it will automatically
+ensure results of math operations where at least one operand is a `CountedFloat` will also be a `CountedFloat`.
+This way we ensure flop counting is a 'closed system'.
+
+On top of this, we monkey-patch the `math` module to ensure that all math operations
+that require counting (`sqrt`, `log2`, `pow`, ...) are also instrumented.
+
+**Example 1**:
+
+```python
+from counted_float import CountedFloat
+
+cf = CountedFloat(1.3)
+f = 2.8
+
+result = cf + f  # result = CountedFloat(4.1)
+
+is_float_1 = isinstance(cf, float)  # True
+is_float_2 = isinstance(result, float)  # True
+```
+
+**Example 2**:
+
+```python
+import math
+from counted_float import CountedFloat
+
+cf1 = CountedFloat(0.81)
+
+s = math.sqrt(cf1)  # s = CountedFloat(0.9)
+is_float = isinstance(s, float)  # True
+```
+
+## 2.2. FLOP counting context managers
+
+Once we use the `CountedFloat` class, we can use the available context managers to count the number of
+flops performed by `CountedFloat` objects.
+
+**Example 1**:  _basic usage_
+```python
+from counted_float import CountedFloat, FlopCountingContext
+
+cf1 = CountedFloat(1.73)
+cf2 = CountedFloat(2.94)
+
+with FlopCountingContext() as ctx:
+    _ = cf1 * cf2
+    _ = cf1 + cf2
+
+counts = ctx.flop_counts()   # {FlopType.MUL: 1, FlopType.ADD: 1}
+counts.total_count()         # 2
+```
+
+**Example 2**:  _pause counting 1_
+
+```python
+from counted_float import CountedFloat, FlopCountingContext
+
+cf1 = CountedFloat(1.73)
+cf2 = CountedFloat(2.94)
+
+with FlopCountingContext() as ctx:
+    _ = cf1 * cf2
+    ctx.pause()
+    _ = cf1 + cf2   # will be executed but not counted
+    ctx.resume()
+    _ = cf1 - cf2
+
+counts = ctx.flop_counts()   # {FlopType.MUL: 1, FlopType.SUB: 1}
+counts.total_count()         # 2
+```
+
+**Example 3**:  _pause counting 2_
+
+```python
+from counted_float import CountedFloat, FlopCountingContext, PauseFlopCounting
+
+cf1 = CountedFloat(1.73)
+cf2 = CountedFloat(2.94)
+
+with FlopCountingContext() as ctx:
+    _ = cf1 * cf2
+    with PauseFlopCounting():
+        _ = cf1 + cf2   # will be executed but not counted
+    _ = cf1 - cf2
+
+counts = ctx.flop_counts()   # {FlopType.MUL: 1, FlopType.SUB: 1}
+counts.total_count()         # 2
+```
+
+## 2.3. Weighted FLOP counting
+
+The `counted_float` package contains a set of default, built-in FLOP weights, based on both empirical measurements
+and theoretical estimates of the relative cost of different floating point operations. 
+
+See [fpu_data_sources.md](https://github.com/bertpl/counted-float/tree/develop/docs/analysis_methodology.md) for
+rationale behind choice of data sources and methodology.
+
+```
+>>> from counted_float.config import get_active_flop_weights
+>>> get_active_flop_weights().show()
+
+{
+    FlopType.ABS        [abs(x)]        :    1
+    FlopType.MINUS      [-x]            :    1
+    FlopType.COMP       [x<=y]          :    1
+    FlopType.RND        [round]         :    2
+    FlopType.F2I        [float->int]    :    2
+    FlopType.I2F        [int->float]    :    2
+    FlopType.ADD        [x+y]           :    1
+    FlopType.SUB        [x-y]           :    1
+    FlopType.MUL        [x*y]           :    1
+    FlopType.DIV        [x/y]           :    5
+    FlopType.SQRT       [sqrt(x)]       :    6
+    FlopType.CBRT       [cbrt(x)]       :   42
+    FlopType.EXP        [e^x]           :   19
+    FlopType.EXP2       [2^x]           :   29
+    FlopType.EXP10      [10^x]          :   23
+    FlopType.LOG        [log(x)]        :   19
+    FlopType.LOG2       [log2(x)]       :   24
+    FlopType.LOG10      [log10(x)]      :   19
+    FlopType.POW        [x^y]           :   62
+    FlopType.SIN        [sin(x)]        :   32
+    FlopType.COS        [cos(x)]        :   31
+    FlopType.TAN        [tan(x)]        :   34
+}
+```
+These weights will be used by default when extracting total weighted flop costs:
+
+```python
+import math
+from counted_float import CountedFloat, FlopCountingContext
+
+
+cf1 = CountedFloat(1.73)
+cf2 = CountedFloat(2.94)
+
+with FlopCountingContext() as ctx:
+    _ = cf1 + cf2
+    _ = cf1 ** cf2
+    _ = math.log2(cf2)
+    
+flop_counts = ctx.flop_counts()
+total_cost = flop_counts.total_weighted_cost()  # 1 + 62 + 24 = 87
+```
+Note that the `total_weighted_cost` method will use the default flop weights as returned by `get_flop_weights()`.  This can be
+overridden by either configuring different flop weights (see next section) or by setting the `weights` argument of the `total_weighted_cost()` method.
+
+
+## 2.4. Configuring FLOP weights
+
+We showed earlier that the `get_flop_weights()` function returns the default FLOP weights.  We can change this by
+using the `set_flop_weights()` function, which takes a `FlopWeights` object as an argument.  This way we can configure
+flop weights that might be obtained using benchmarks run on the target hardware (see later sections).
+
+```python
+from counted_float.config import set_active_flop_weights
+from counted_float import FlopWeights
+
+set_active_flop_weights(weights=FlopWeights(...))  # insert own weights here
+```
+## 2.5. Inspecting built-in data
+
+### 2.5.1. Default, pre-aggregated flop weights
+
+Built-in flop weights can be inspected using the following functions:
+
+```python
+from counted_float.config import get_default_consensus_flop_weights
+
+>>> get_default_consensus_flop_weights(rounded=False).show()
+
+{
+    FlopType.ABS        [abs(x)]        :   0.63673
+    FlopType.MINUS      [-x]            :   0.64396
+    FlopType.COMP       [x<=y]          :   1.20756
+    FlopType.RND        [round]         :   1.54041
+    FlopType.F2I        [float->int]    :   1.99099
+    FlopType.I2F        [int->float]    :   1.84601
+    FlopType.ADD        [x+y]           :   1.00000
+    FlopType.SUB        [x-y]           :   1.00586
+    FlopType.MUL        [x*y]           :   1.37238
+    FlopType.DIV        [x/y]           :   5.07465
+    FlopType.SQRT       [sqrt(x)]       :   5.90559
+    FlopType.CBRT       [cbrt(x)]       :  42.39375
+    FlopType.EXP        [e^x]           :  18.58228
+    FlopType.EXP2       [2^x]           :  28.88672
+    FlopType.EXP10      [10^x]          :  22.86839
+    FlopType.LOG        [log(x)]        :  18.89135
+    FlopType.LOG2       [log2(x)]       :  24.34792
+    FlopType.LOG10      [log10(x)]      :  18.55085
+    FlopType.POW        [x^y]           :  61.79155
+    FlopType.SIN        [sin(x)]        :  31.91490
+    FlopType.COS        [cos(x)]        :  30.79295
+    FlopType.TAN        [tan(x)]        :  34.37970
+}
+```
+
+The default weights that are configured out-of-the-box in the package are the integer-rounded `consensus` weights.
+
+### 2.5.2. Custom-aggregated flop weights
+
+We can retrieve built-in flop weights in a more fine-grained manner, by custom filtering and the aggregating them with
+the geometric mean.
+
+```python
+from counted_float.config import get_builtin_flop_weights
+
+>>> get_builtin_flop_weights(key_filter="arm").show()
+
+{
+    FlopType.ABS        [abs(x)]        :   0.97313
+    FlopType.MINUS      [-x]            :   0.99098
+    FlopType.COMP       [x<=y]          :   1.03987
+    FlopType.RND        [round]         :   1.35111
+    FlopType.F2I        [float->int]    :   1.52648
+    FlopType.I2F        [int->float]    :   1.63320
+    FlopType.ADD        [x+y]           :   1.00000
+    FlopType.SUB        [x-y]           :   1.00058
+    FlopType.MUL        [x*y]           :   1.44952
+    FlopType.DIV        [x/y]           :   5.00897
+    FlopType.SQRT       [sqrt(x)]       :   5.15597
+    FlopType.CBRT       [cbrt(x)]       :  39.30448
+    FlopType.EXP        [e^x]           :  17.22817
+    FlopType.EXP2       [2^x]           :  15.82232
+    FlopType.EXP10      [10^x]          :  21.20195
+    FlopType.LOG        [log(x)]        :  17.51472
+    FlopType.LOG2       [log2(x)]       :  18.32529
+    FlopType.LOG10      [log10(x)]      :  17.19903
+    FlopType.POW        [x^y]           :  47.63289
+    FlopType.SIN        [sin(x)]        :  29.58923
+    FlopType.COS        [cos(x)]        :  28.54904
+    FlopType.TAN        [tan(x)]        :  31.87442
+}
+```
+
+# 3. Benchmarking
+
+If the package is installed with the optional `numba` dependency, it provides the ability to micro-benchmark 
+floating point operations as follows:
+
+```
+>>> from counted_float.benchmarking import run_flops_benchmark
+>>> results = run_flops_benchmark()
+
+Running FLOPS benchmarks using counted-float 0.9.5 ...
+(Expected duration: ~87.8 seconds)
+
+baseline                           : wwwwwwwwwwwwwww.........................   [  74.43 ns ±  2.6% |   302 cpu cycles ±  2.6% ]  /  1000 iterations
+add                                : wwwwwwwwwwwwwww.........................   [ 662.35 ns ±  0.2% | 2.69K cpu cycles ±  0.2% ]  /  1000 iterations
+add_minus                          : wwwwwwwwwwwwwww.........................   [   1.23 µs ±  0.2% | 4.98K cpu cycles ±  0.2% ]  /  1000 iterations
+add_abs                            : wwwwwwwwwwwwwww.........................   [   1.23 µs ±  0.4% | 4.99K cpu cycles ±  0.4% ]  /  1000 iterations
+add_add                            : wwwwwwwwwwwwwww.........................   [   1.29 µs ±  0.2% | 5.23K cpu cycles ±  0.2% ]  /  1000 iterations
+add_sub                            : wwwwwwwwwwwwwww.........................   [   1.29 µs ±  0.2% | 5.23K cpu cycles ±  0.2% ]  /  1000 iterations
+add_round                          : wwwwwwwwwwwwwww.........................   [   1.44 µs ±  0.1% | 5.84K cpu cycles ±  0.1% ]  /  1000 iterations
+add_sqrt                           : wwwwwwwwwwwwwww.........................   [   3.96 µs ±  0.2% | 16.1K cpu cycles ±  0.2% ]  /  1000 iterations
+add_cbrt                           : wwwwwwwwwwwwwww.........................   [  25.42 µs ±  0.2% |  103K cpu cycles ±  0.2% ]  /  1000 iterations
+add_log                            : wwwwwwwwwwwwwww.........................   [  11.69 µs ±  0.3% | 47.4K cpu cycles ±  0.3% ]  /  1000 iterations
+add_log_exp                        : wwwwwwwwwwwwwww.........................   [  22.57 µs ±  0.1% | 91.5K cpu cycles ±  0.1% ]  /  1000 iterations
+add_log2                           : wwwwwwwwwwwwwww.........................   [  12.00 µs ±  0.2% | 48.7K cpu cycles ±  0.2% ]  /  1000 iterations
+add_log2_exp2                      : wwwwwwwwwwwwwww.........................   [  22.48 µs ±  0.2% | 91.2K cpu cycles ±  0.2% ]  /  1000 iterations
+add_log10                          : wwwwwwwwwwwwwww.........................   [  11.50 µs ±  0.2% | 46.6K cpu cycles ±  0.2% ]  /  1000 iterations
+add_log10_exp10                    : wwwwwwwwwwwwwww.........................   [  24.68 µs ±  0.2% |  100K cpu cycles ±  0.2% ]  /  1000 iterations
+add_sin                            : wwwwwwwwwwwwwww.........................   [  18.64 µs ±  0.3% | 75.6K cpu cycles ±  0.3% ]  /  1000 iterations
+add_cos                            : wwwwwwwwwwwwwww.........................   [  18.92 µs ±  0.3% | 76.7K cpu cycles ±  0.3% ]  /  1000 iterations
+add_tan                            : wwwwwwwwwwwwwww.........................   [  20.91 µs ±  0.2% | 84.8K cpu cycles ±  0.2% ]  /  1000 iterations
+pow                                : wwwwwwwwwwwwwww.........................   [  24.12 µs ±  0.3% | 97.8K cpu cycles ±  0.3% ]  /  1000 iterations
+pow_pow                            : wwwwwwwwwwwwwww.........................   [  48.15 µs ±  0.2% |  195K cpu cycles ±  0.2% ]  /  1000 iterations
+sub                                : wwwwwwwwwwwwwww.........................   [ 661.55 ns ±  0.2% | 2.68K cpu cycles ±  0.2% ]  /  1000 iterations
+sub_sub                            : wwwwwwwwwwwwwww.........................   [   1.29 µs ±  0.2% | 5.24K cpu cycles ±  0.2% ]  /  1000 iterations
+mul                                : wwwwwwwwwwwwwww.........................   [ 961.78 ns ±  0.2% | 3.90K cpu cycles ±  0.2% ]  /  1000 iterations
+mul_mul                            : wwwwwwwwwwwwwww.........................   [   1.92 µs ±  0.2% | 7.78K cpu cycles ±  0.2% ]  /  1000 iterations
+div                                : wwwwwwwwwwwwwww.........................   [   2.45 µs ±  0.2% | 9.92K cpu cycles ±  0.2% ]  /  1000 iterations
+div_div                            : wwwwwwwwwwwwwww.........................   [   5.00 µs ±  0.2% | 20.3K cpu cycles ±  0.2% ]  /  1000 iterations
+lte_addsub                         : wwwwwwwwwwwwwww.........................   [   1.71 µs ±  0.2% | 6.94K cpu cycles ±  0.2% ]  /  1000 iterations
+
+>>> results.flop_weights.show() 
+
+{
+    FlopType.ABS        [abs(x)]        :   0.90556
+    FlopType.MINUS      [-x]            :   0.90089
+    FlopType.COMP       [x<=y]          :   1.67297
+    FlopType.RND        [round]         :   1.24118
+    FlopType.ADD        [x+y]           :   1.00000
+    FlopType.SUB        [x-y]           :   0.99928
+    FlopType.MUL        [x*y]           :   1.52656
+    FlopType.DIV        [x/y]           :   4.06589
+    FlopType.SQRT       [sqrt(x)]       :   5.26487
+    FlopType.CBRT       [cbrt(x)]       :  39.49190
+    FlopType.EXP        [e^x]           :  17.34508
+    FlopType.EXP2       [2^x]           :  16.70475
+    FlopType.EXP10      [10^x]          :  21.03351
+    FlopType.LOG        [log(x)]        :  17.59412
+    FlopType.LOG2       [log2(x)]       :  18.08932
+    FlopType.LOG10      [log10(x)]      :  17.27955
+    FlopType.POW        [x^y]           :  38.32044
+    FlopType.SIN        [sin(x)]        :  28.67006
+    FlopType.COS        [cos(x)]        :  29.11818
+    FlopType.TAN        [tan(x)]        :  32.28703
+    FlopType.F2I        [float->int]    :       nan
+    FlopType.I2F        [int->float]    :       nan
+}
+```
+
+## 4. Installing the package as a command-line tool
+
+An alternative way of using (parts) of the functionality is installing the package as a stand-alone command-line tool
+using `uv` or `pipx`:
+
+```
+uv tool install git+https://github.com/bertpl/counted-float@main[numba,cli]         # latest official release
+uv tool install git+https://github.com/bertpl/counted-float@develop[numba,cli]      # or latest develop version
+```
+This installs the `counted_float` command-line tool, which can be used to e.g. run flops benchmarks.
+
+## 4.1. Running benchmarks
+
+```
+counted_float benchmark
+```
+after which the results will be shown as .json.
+
+## 4.2. Show built-in data
+
+```
+[~] counted_float show-data
+                                                              MINUS       ABS      COMP       SUB       ADD       MUL       RND       F2I       I2F       DIV      SQRT      EXP2       EXP       LOG      LOG2     EXP10     LOG10       SIN       COS       POW       TAN      CBRT
+ALL                                                            0.47      0.70      0.97      1.00      1.00      1.40      1.77      1.91      1.92      5.53      7.36     15.86     17.52     19.00     22.38     23.03     24.66     30.41     31.39     41.81     42.16     44.32
+ ├─arm                                                         0.89      1.05      0.64      1.01      1.00      1.50      1.61      1.49      1.59      6.15      7.60     15.98     18.57     19.60     20.86     23.30     24.26     32.32     34.08     42.28     42.85     44.97
+ │  ├─v8_x                                                     0.83      1.00      0.67      1.01      1.00      1.48      1.50      1.56      1.91      5.90      7.42     15.45     17.86     19.06     20.80     22.92     22.90     32.43     33.47     41.19     42.99     44.71
+ │  │  ├─benchmarks                                            0.70      1.01      0.45      1.01      1.00      1.46      1.50        /         /       4.65      6.68     13.93     16.10     17.18     18.75     20.66     20.65     29.23     30.17     37.13     38.76     40.30
+ │  │  │  ├─apple_m3_max_mbp16                                 0.90      0.90      1.64      1.00      1.00      1.40      1.25        /         /       4.00      5.28     16.69     17.42     17.59     18.15     21.06     17.31     28.77     29.17     37.93     39.04     39.48
+ │  │  │  ├─apple_m3_mba15                                     0.90      0.90      1.68      1.00      1.00      1.53      1.24        /         /       4.06      5.27     16.68     17.43     17.56     18.09     21.07     17.29     28.63     29.06     38.46     32.32     39.43
+ │  │  │  ├─ec2_m6g_xlarge_graviton_2_neoverse_n1              0.46      1.47      0.38      1.00      1.00      1.38      2.01        /         /       5.42      8.76     10.21     14.17     15.98     17.91     22.65     25.25     28.68     29.63     33.95     41.90     40.25
+ │  │  │  ├─ec2_m7g_xlarge_graviton_3_neoverse_v1              0.51      1.00      0.01      1.00      1.00      1.50      1.99        /         /       6.03      9.58     11.89     14.81     17.49     19.20     19.09     27.67     30.26     31.49     36.85     42.45     40.96
+ │  │  │  └─gh_apple_m1                                        0.86      0.86      1.73      1.06      1.00      1.48      1.22        /         /       4.07      5.70     15.50     16.98     17.33     20.52     19.61     17.95     29.86     31.59     38.67     38.97     41.44
+ │  │  └─specs                                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.73      2.12      7.50      8.25        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v8_cortex_a76                                  1.00      1.00      1.00      1.00      1.00      1.50      1.50      2.00      3.00      7.50      8.50        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v9_cortex_x1                                   1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v9_neoverse_n1                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      2.00      3.00      7.50      8.50        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     └─arm_v9_neoverse_v1                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  ├─v9_0                                                     0.74      1.00      0.25      1.00      1.00      1.51      1.80      1.27      1.27      6.78      8.83     14.32     17.90     20.66     22.09     24.04     32.41     35.53     37.31     44.86     52.70     48.30
+ │  │  ├─benchmarks                                            0.54      1.00      0.06      1.00      1.00      1.52      2.16        /         /       6.14      9.75     12.12     15.15     17.49     18.70     20.35     27.44     30.08     31.59     37.98     44.62     40.89
+ │  │  │  ├─ec2_m8g_xlarge_graviton_4_neoverse_v2              0.52      1.00      0.01      1.00      1.00      1.50      2.05        /         /       6.04      9.49     12.66     14.40     17.66     18.54     19.86     27.12     28.76     30.11     36.90     42.70     38.65
+ │  │  │  └─gh_azure_cobalt_100_neoverse_n2                    0.57      1.00      0.33      1.00      1.00      1.53      2.28        /         /       6.23     10.02     11.61     15.94     17.33     18.87     20.85     27.76     31.46     33.15     39.09     46.62     43.26
+ │  │  └─specs                                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v9_cortex_x2                                   1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v9_cortex_x3                                   1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     ├─arm_v9_neoverse_n2                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  │     └─arm_v9_neoverse_v2                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      7.50      8.00        /         /         /         /         /         /         /         /         /         /         / 
+ │  └─v9_2                                                     1.16      1.16      1.55      1.02      1.00      1.51      1.55      1.66      1.66      5.80      6.70     18.44     20.04     19.11     19.75     22.96     19.23     29.29     31.69     40.89     34.73     42.11
+ │     ├─benchmarks                                            1.35      1.35      2.40      1.03      1.00      1.51      1.84        /         /       5.32      7.10     20.38     22.15     21.13     21.83     25.38     21.25     32.38     35.03     45.20     38.39     46.55
+ │     │  └─apple_m4_pro_mbp16                                 1.35      1.35      2.40      1.03      1.00      1.51      1.84        /         /       5.32      7.10     20.38     22.15     21.13     21.83     25.38     21.25     32.38     35.03     45.20     38.39     46.55
+ │     └─specs                                                 1.00      1.00      1.00      1.00      1.00      1.50      1.31      1.50      1.50      6.33      6.33        /         /         /         /         /         /         /         /         /         /         / 
+ │        ├─arm_v9_cortex_x4                                   1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      6.50      6.50        /         /         /         /         /         /         /         /         /         /         / 
+ │        ├─arm_v9_cortex_x925                                 1.00      1.00      1.00      1.00      1.00      1.50      1.00      1.50      1.50      6.00      6.00        /         /         /         /         /         /         /         /         /         /         / 
+ │        └─arm_v9_neoverse_v3                                 1.00      1.00      1.00      1.00      1.00      1.50      1.50      1.50      1.50      6.50      6.50        /         /         /         /         /         /         /         /         /         /         / 
+ └─x86                                                         0.24      0.47      1.47      0.99      1.00      1.31      1.95      2.47      2.32      4.98      7.12     15.74     16.52     18.43     24.01     22.76     25.06     28.61     28.92     41.35     41.47     43.69
+    ├─amd                                                      0.29      0.43      1.51      0.99      1.00      1.19      1.21      2.42      2.54      4.74      7.39     18.48     16.73     18.23     24.93     24.05     24.46     28.42     28.69     42.99     40.87     45.73
+    │  ├─2017_zen1                                             0.34      0.34      1.16      1.00      1.00      1.29      1.34      3.07      3.07      4.18      6.44     63.95     19.71     17.05     32.24     43.83     18.79     30.58     30.27     57.99     39.24     66.55
+    │  │  ├─analysis_uops_info_zen1+                           0.33      0.33      2.33      1.00      1.00      1.33      1.33      3.33      3.33      4.33      6.67        /         /         /         /         /         /         /         /         /         /         / 
+    │  │  └─benchmark_ryzen_1700x                              0.35      0.35      0.58      1.01      1.00      1.24      1.34        /         /       4.03      6.23     58.90     18.16     15.70     29.69     40.37     17.30     28.16     27.88     53.41     36.14     61.29
+    │  ├─2020_zen3                                             0.18      0.34      1.16      1.00      1.00      1.03      1.00      2.15      2.33      4.36      6.76     11.76     15.94     17.09     22.60     19.44     21.26     24.19     24.21     38.36     38.26     39.30
+    │  │  ├─analysis_agner_fog_r7_5800x                        0.33      0.33      1.67      1.00      1.00        /       1.00      2.00      2.33      4.50      6.67        /         /         /         /         /         /         /         /         /         /         / 
+    │  │  ├─analysis_uops_info_zen3                            0.33      0.33      2.33      1.00      1.00      1.00      1.00      3.00      3.00      4.33      6.67        /         /         /         /         /         /         /         /         /         /         / 
+    │  │  ├─benchmark_gh_amd_epyc_7763_linux                   0.10      0.34      0.68      1.00      1.00      1.00      1.00        /         /       4.31      6.86     10.35     11.79     15.35     14.36     14.53     22.21     22.45     22.86     30.57     33.34     31.25
+    │  │  └─benchmark_gh_amd_epyc_7763_windows                 0.10      0.34      0.68      1.00      1.00      1.00      1.00        /         /       4.28      6.86     10.32     16.65     14.70     27.48     20.08     15.72     20.14     19.82     37.19     33.92     38.17
+    │  ├─2022_zen4                                             0.23      0.33      1.34      0.99      1.00      1.01      1.00      1.92      1.85      4.37      6.90     11.20     12.97     17.81     21.31     16.47     26.04     25.08     25.55     33.51     37.44     34.67
+    │  │  ├─analysis_agner_fog_r9_7900x                        0.33      0.33      1.33      1.00      1.00        /       1.00      2.00      2.00      4.33      7.00        /         /         /         /         /         /         /         /         /         /         / 
+    │  │  ├─analysis_uops_info_zen4                            0.33      0.33      2.33      1.00      1.00      1.00      1.00      3.00      2.67      4.33      7.00        /         /         /         /         /         /         /         /         /         /         / 
+    │  │  ├─benchmark_ec2_m7a_xlarge_zen4                      0.07      0.33      1.57      0.97      1.00      0.98      0.99        /         /       4.49      6.95      9.97     11.55     15.86     18.98     14.66     23.19     22.34     22.75     29.84     33.34     30.87
+    │  │  └─specs_amd                                          0.33      0.33      0.67      1.00      1.00      1.00      1.00      1.33      1.33      4.33      6.67        /         /         /         /         /         /         /         /         /         /         / 
+    │  └─2024_zen5                                             0.47      0.91      2.87      0.97      1.00      1.51      1.62      2.69      3.14      6.37      9.89     13.84     19.21     21.29     24.85     23.87     34.44     35.17     36.16     45.83     49.62     48.23
+    │     ├─analysis_agner_fog_r7_9800x3d                      1.00      1.00      3.00      1.00      1.00        /       1.50      3.00      3.50      6.50     10.00        /         /         /         /         /         /         /         /         /         /         / 
+    │     ├─benchmark_ec2_m8a_xlarge_zen5                      0.21      1.53      2.95      0.91      1.00      1.36      1.87        /         /       6.12      9.69     13.33     18.50     20.50     23.94     22.98     33.17     33.87     34.83     44.14     47.79     46.45
+    │     └─specs_amd                                          0.50      0.50        /       1.00      1.00      1.50      1.50        /         /       6.50     10.00        /         /         /         /         /         /         /         /         /         /         / 
+    └─intel                                                    0.21      0.50      1.43      0.99      1.00      1.44      3.13      2.51      2.12      5.22      6.86     13.40     16.32     18.63     23.13     21.53     25.68     28.79     29.16     39.77     42.08     41.74
+       ├─2017_coffee_lake_gen_8                                0.14      0.40      1.30      0.98      1.00      1.03      1.99      1.66      1.66      3.54      4.56     10.44     16.58     12.62     19.41     21.85     13.05     20.81     20.80     33.18     28.66     36.32
+       │  ├─analysis_agner_fog_coffee_lake                     0.25      0.25        /       1.00      1.00      1.00      2.00      1.50      1.50      3.50      4.00        /         /         /         /         /         /         /         /         /         /         / 
+       │  ├─analysis_uops_info_coffee_lake                     0.25      0.25      0.75      1.00      1.00      1.00      2.00      1.75      1.75      3.75      4.75        /         /         /         /         /         /         /         /         /         /         / 
+       │  ├─benchmark_gh_intel_i7_8550U_windows                0.08      0.44      1.69      1.01      1.00      1.01      2.00        /         /       3.15      4.79      7.02     15.44     11.99     30.07     21.33     13.00     17.80     17.21     32.36     28.09     35.06
+       │  └─benchmark_gh_intel_i7_8700B_macos                  0.08      0.91      1.77      0.92      1.00      1.10      1.97        /         /       3.78      4.77     16.30     18.68     13.93     13.15     23.47     13.74     25.51     26.38     35.70     30.68     39.48
+       ├─2019_sunny_cove_gen_10                                0.16      0.29      0.78      0.98      1.00      0.99      1.97      1.63      1.63      3.62      4.51     12.52      9.62     14.29     17.52     13.66     20.63     19.62     20.22     29.29     30.32     29.53
+       │  ├─analysis_agner_fog_ice_lake                        0.25      0.25      0.50      1.00      1.00        /       2.00      1.50      1.50      3.50      4.00        /         /         /         /         /         /         /         /         /         /         / 
+       │  ├─analysis_uops_info_ice_lake                        0.25      0.25      0.75      1.00      1.00      1.00      2.00      1.75      1.75      3.75      4.75        /         /         /         /         /         /         /         /         /         /         / 
+       │  ├─analysis_uops_info_tiger_lake                      0.25      0.25      0.75      1.00      1.00      1.00      2.00      1.75      1.75      3.75      4.75        /         /         /         /         /         /         /         /         /         /         / 
+       │  └─benchmark_ec2_m6i_xlarge_ice_lake                  0.04      0.44      1.30      0.91      1.00      1.00      1.89        /         /       3.50      4.60     11.82      9.08     13.49     16.55     12.90     19.48     18.52     19.09     27.65     28.63     27.88
+       ├─2021_golden_cove_gen_12                               0.28      0.55      1.56      0.99      1.00      1.74      3.77      3.02      2.55      6.20      8.22     14.56     17.86     21.33     24.92     23.36     31.87     33.78     34.10     42.83     49.32     45.31
+       │  ├─analysis_uops_info_alder_lake_p                    0.33      0.33      1.00      1.00      1.00      1.33      2.67      2.33      2.33      5.00      6.33        /         /         /         /         /         /         /         /         /         /         / 
+       │  ├─benchmark_ec2_m7i_xlarge_sapphire_rapids           0.13      0.99      2.52      0.98      1.00      1.99      5.03        /         /       6.80      9.74     16.28     19.97     23.84     27.86     26.11     35.63     37.76     38.12     47.88     55.14     50.65
+       │  └─specs_intel                                        0.50      0.50      1.50      1.00      1.00      2.00      4.00      3.50      2.50      7.00      9.00        /         /         /         /         /         /         /         /         /         /         / 
+       ├─2022_raptor_cove_gen_13_14                            0.26      0.71      1.96      1.00      1.00      2.00      4.51      3.55      2.53      7.00      9.49     15.73     19.93     24.00     28.64     26.02     35.73     37.86     38.33     49.02     55.79     51.16
+       │  ├─benchmark_ec2_i7i_xlarge_emerald_rapids            0.13      1.00      2.55      1.00      1.00      2.00      5.08        /         /       7.00     10.00     15.95     20.20     24.33     29.03     26.37     36.21     38.37     38.85     49.69     56.55     51.86
+       │  └─specs_intel                                        0.50      0.50      1.50      1.00      1.00      2.00      4.00      3.50      2.50      7.00      9.00        /         /         /         /         /         /         /         /         /         /         / 
+       └─2023_redwood_cove_ultra_1                             0.24      0.71      1.95      0.99      1.00      1.73      4.51      3.46      2.47      6.97      9.47     14.44     20.41     24.32     27.29     25.53     36.40     37.92     38.32     48.75     55.19     50.95
+          ├─benchmark_ec2_m8i_xlarge_granite_rapids            0.12      1.01      2.53      0.98      1.00      1.50      5.09        /         /       6.94      9.96     14.28     20.19     24.06     26.99     25.26     36.01     37.51     37.91     48.22     54.60     50.40
+          └─specs_intel                                        0.50      0.50      1.50      1.00      1.00      2.00      4.00      3.50      2.50      7.00      9.00        /         /         /         /         /         /         /         /         /         /         / 
+ 
+```
+
+# 5. Known limitations
+
+- currently any non-Python-built-in math operations are not counted (e.g. `numpy`)
+- not all Python built-in math operations are counted (e.g. `log`, `log10`, `exp`, `exp10`)
+- flop weights should be taken with a grain of salt and should only provide relative ballpark estimates w.r.t computational complexity.  Production implementations in a compiled language could have vastly differing performance depending on cpu cache sizes, branch prediction misses, compiler optimizations using vector operations (AVX etc...), etc...
+ 
+# Appendix A - Flop counting / analysis details
+
+This appendix provides detailed information about how each floating-point operation (FLOP) type is counted and analyzed in the `counted-float` package. For each flop type, you will find:
+- Relevant scalar instructions for ARM (v8+) and x86 (SSE2+)
+- Python operations that are counted for this flop type
+- Python operations that are *not* counted for this flop type
+
+## Flop Types
+
+### FlopType.ABS (`abs(x)`)
+- Relevant CPU instructions
+  - **ARM:** `FABS`
+  - **x86:** `ANDPD`
+- **Counted Python operations:** `abs(x)` where `x` is a `CountedFloat`
+- **Not counted:** `numpy.abs`, complex abs, abs on non-CountedFloat
+
+### FlopType.MINUS (`-x`)
+- Relevant CPU instructions
+  - **ARM:** `FNEG`
+  - **x86:** `XORPD`
+- **Counted Python operations:** Unary minus (`-x`) for `CountedFloat`
+- **Not counted:** Negation on non-CountedFloat, numpy negation
+
+### FlopType.COMP (`x<=y`, `x>y`, `x==y`, `x==0.0`, ...)
+- Relevant CPU instructions
+  - **ARM:** `FCMP`
+  - **x86:** `(U)COMISD`
+- **Counted Python operations:** `x == y`, `x != y`, `x <= y`, ... and `min(x,y)`, `max(x,y)` for `CountedFloat`
+- **Not counted:** Comparisons on non-CountedFloat, numpy comparisons
+
+### FlopType.RND (`round`)
+- Relevant CPU instructions
+  - **ARM:** `FRINT`
+  - **x86:** `ROUNDSD`
+- **Counted Python operations:** `round(x, 0)` for `CountedFloat` (returns float)
+- **Not counted:** `numpy.round`, rounding with decimals, rounding on non-CountedFloat
+
+### FlopType.F2I (`float->int`)
+- Relevant CPU instructions
+  - **ARM:** `FCVTZS`
+  - **x86:** `CVTSD2SI`
+- **Counted Python operations:** `int(x)`, `math.floor(x)`, `math.ceil(x)`, `math.trunc(x)`, `round(x)` for `CountedFloat` (returns int)
+- **Not counted:** Conversions on non-CountedFloat, numpy conversions
+
+### FlopType.I2F (`int->float`)
+- Relevant CPU instructions
+  - **ARM:** `SCVTF`
+  - **x86:** `CVTSI2SD`
+- **Counted Python operations:** Construction of `CountedFloat` from int, any binary operation where one operand is an int and the other a `CountedFloat` (e.g., `x + 3`, `3 * x`, etc.)
+- **Not counted:** `float(n)`, unitary operations (e.g. `math.sqrt` on integers -> convert to `CountedFloat` first)
+
+### FlopType.ADD (`x+y`)
+- Relevant CPU instructions
+  - **ARM:** `FADD`
+  - **x86:** `ADDSD`
+- **Counted Python operations:** `x + y` or `y + x` for `CountedFloat`
+- **Not counted:** Addition on non-CountedFloat, numpy addition
+
+### FlopType.SUB (`x-y`)
+- Relevant CPU instructions
+  - **ARM:** `FSUB`
+  - **x86:** `SUBSD`
+- **Counted Python operations:** `x - y` or `y - x` for `CountedFloat`
+- **Not counted:** Subtraction on non-CountedFloat, numpy subtraction
+
+### FlopType.MUL (`x*y`)
+- Relevant CPU instructions
+  - **ARM:** `FMUL`
+  - **x86:** `MULSD`
+- **Counted Python operations:** `x * y` or `y * x` for `CountedFloat`
+- **Not counted:** Multiplication on non-CountedFloat, numpy multiplication
+
+### FlopType.DIV (`x/y`)
+- Relevant CPU instructions
+  - **ARM:** `FDIV`
+  - **x86:** `DIVSD`
+- **Counted Python operations:** `x / y` or `y / x` for `CountedFloat`
+- **Not counted:** Division on non-CountedFloat, numpy division
+
+### FlopType.SQRT (`sqrt(x)`)
+- Relevant CPU instructions
+  - **ARM:** `FSQRT`
+  - **x86:** `SQRTSD`
+- **Counted Python operations:** `math.sqrt(x)` for `CountedFloat`
+- **Not counted:** `numpy.sqrt`, sqrt on non-CountedFloat
+
+### FlopType.CBRT (`cbrt(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.cbrt(x)` for `CountedFloat`
+- **Not counted:** `numpy.cbrt`, cbrt on non-CountedFloat
+
+### FlopType.EXP (`e^x`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.exp(x)` for `CountedFloat`
+- **Not counted:** `math.exp(x)` on non-CountedFloat, `numpy.exp`, `math.expm1`, `math.e ** x`
+
+### FlopType.EXP2 (`2^x`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `2 ** x`, `pow(2, x)` or `math.exp2(x)` for `CountedFloat`
+- **Not counted:** `exp2` on non-CountedFloat, `numpy.exp2`
+
+### FlopType.EXP10 (`10^x`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `10 ** x`, `pow(10, x)` for `CountedFloat`
+- **Not counted:** `10 ** x` on non-CountedFloat
+
+### FlopType.LOG (`log(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.log(x)` for `CountedFloat`
+- **Not counted:** `numpy.log`, log on non-CountedFloat
+
+### FlopType.LOG2 (`log2(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.log2(x)` for `CountedFloat`
+- **Not counted:** `numpy.log2`, log2 on non-CountedFloat
+
+### FlopType.LOG10 (`log10(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.log10(x)` for `CountedFloat`
+- **Not counted:** `numpy.log10`, log10 on non-CountedFloat
+
+### FlopType.POW (`x^y`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `x ** y`, `pow(x, y)` for `CountedFloat`
+- **Not counted:** `pow` on non-CountedFloat, `numpy.pow`
+
+### FlopType.SIN (`sin(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.sin(x)` for `CountedFloat`
+- **Not counted:** `sin` on non-CountedFloat, `numpy.sin`
+
+### FlopType.COS (`cos(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.cos(x)` for `CountedFloat`
+- **Not counted:** `cos` on non-CountedFloat, `numpy.cos`
+
+### FlopType.TAN (`tan(x)`)
+- Relevant CPU instructions
+  - **ARM:** (software)
+  - **x86:** (software)
+- **Counted Python operations:** `math.tan(x)` for `CountedFloat`
+- **Not counted:** `tan` on non-CountedFloat, `numpy.tan