Optimise getValueOfBits and insertBits with BMI2 PEXT/PDEP (#717) #796
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,3 +1,15 @@ | ||
| # @author Tyson Jones | ||
|
|
||
| add_all_local_examples() | ||
|
|
||
| # The issue-#717 bitwise micro-benchmark builds with -mbmi2 (so its PEXT/PDEP path is enabled) only | ||
| # when the user opts in via QUEST_ENABLE_BMI2 — same switch the library uses. Without the opt-in it | ||
| # compiles the scalar fallback and prints "BMI2 fast path: INACTIVE" (never SIGILLs). add_example() | ||
| # names the target <filename>_<ext>; the flag is scoped to this one target. | ||
| if (QUEST_ENABLE_BMI2 AND TARGET benchmark_bitwise_bmi2_cpp) | ||
| include(CheckCXXCompilerFlag) | ||
| check_cxx_compiler_flag("-mbmi2" QUEST_EXAMPLE_SUPPORTS_MBMI2) | ||
| if (QUEST_EXAMPLE_SUPPORTS_MBMI2) | ||
| target_compile_options(benchmark_bitwise_bmi2_cpp PRIVATE -mbmi2) | ||
| endif() | ||
| endif() |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,153 @@ | ||
| /** @file | ||
| * A quick, self-contained micro-benchmark of the BMI2 PEXT/PDEP fast paths added for issue #717, | ||
| * comparing them against the original scalar bit gather/scatter loops. It prints per-call timings | ||
| * so QuEST's CI can compare the speedup across its tested platforms and compilers. | ||
| * | ||
| * The two scalar routines below mirror getValueOfBits() and insertBitsWithMaskedValues() from | ||
| * quest/src/core/bitwise.hpp; the BMI2 routines are the single-instruction _pext_u64 / _pdep_u64 | ||
| * paths. This file deliberately depends on nothing but the C++ standard library (and <immintrin.h> | ||
| * when targeting x86 BMI2), so it compiles and runs on every platform — emitting the scalar | ||
| * timings alone where BMI2 is unavailable, never raising SIGILL. | ||
| * | ||
| * Build note: this target is compiled with -mbmi2 (see examples/automated/CMakeLists.txt) so the | ||
| * intrinsic path is enabled; the QuEST library itself enables -mbmi2 the same way in the top-level | ||
| * CMakeLists.txt. Whether the fast path was compiled in is printed at runtime. | ||
| * | ||
| * @author (issue #717 contribution) | ||
| */ | ||
|
|
||
| #include <cstdint> | ||
| #include <cstdio> | ||
| #include <chrono> | ||
|
|
||
| #if defined(__BMI2__) && (defined(__x86_64__) || defined(__i386__) || defined(_M_X64) || defined(_M_IX86)) | ||
| #include <immintrin.h> | ||
| #define BENCH_USE_BMI2 | ||
| #endif | ||
|
|
||
| using std::uint64_t; | ||
|
|
||
| // --- scalar references (mirroring quest/src/core/bitwise.hpp) ------------------------------------- | ||
|
|
||
| // getValueOfBits: gather the bits at the given (strictly increasing) positions into the low bits. | ||
| static inline uint64_t scalarGather(uint64_t number, const int* inds, int n) { | ||
| uint64_t value = 0; | ||
| for (int i=0; i<n; i++) | ||
| value |= ((number >> inds[i]) & 1ULL) << i; | ||
| return value; | ||
| } | ||
|
|
||
| // insertBitsWithMaskedValues: spread number's low bits into the positions NOT named by inds (i.e. | ||
| // insert a 0 at each increasing index), then OR in the precomputed value mask. | ||
| static inline uint64_t scalarScatter(uint64_t number, const int* inds, int n, uint64_t valueMask) { | ||
| uint64_t r = number; | ||
| for (int i=0; i<n; i++) { | ||
| uint64_t lo = r & ((1ULL << inds[i]) - 1); | ||
| uint64_t hi = r & ~((1ULL << inds[i]) - 1); | ||
| r = (hi << 1) | lo; | ||
| } | ||
| return valueMask | r; | ||
| } | ||
|
|
||
| static inline uint64_t makePosMask(const int* inds, int n) { | ||
| uint64_t m = 0; | ||
| for (int i=0; i<n; i++) | ||
| m |= 1ULL << inds[i]; | ||
| return m; | ||
| } | ||
|
|
||
| // --- timing harness ------------------------------------------------------------------------------ | ||
|
|
||
| static double nsPerCall(uint64_t iters, double seconds) { | ||
| return 1e9 * seconds / (double) iters; | ||
| } | ||
|
|
||
| template <typename F> | ||
| static double timeMin(uint64_t iters, int reps, F&& fn) { | ||
| double best = 1e300; | ||
| for (int r=0; r<reps; r++) { | ||
| auto t0 = std::chrono::steady_clock::now(); | ||
| fn(iters); | ||
| auto t1 = std::chrono::steady_clock::now(); | ||
| double s = std::chrono::duration<double>(t1 - t0).count(); | ||
| if (s < best) best = s; | ||
| } | ||
| return best; | ||
| } | ||
|
|
||
| int main() { | ||
|
|
||
| printf("QuEST issue #717 - BMI2 PEXT/PDEP bitwise micro-benchmark\n"); | ||
| #ifdef BENCH_USE_BMI2 | ||
| printf("BMI2 fast path: ACTIVE (compiled with -mbmi2)\n\n"); | ||
| #else | ||
| printf("BMI2 fast path: INACTIVE (x86 BMI2 not targeted; scalar timings only)\n\n"); | ||
| #endif | ||
|
|
||
| const uint64_t iters = 8000000; // keeps total runtime well under a second | ||
| const int reps = 3; | ||
| const int counts[] = {3, 6}; // representative qubit-arity per gate | ||
|
|
||
| printf("%-8s %-4s %14s %14s %10s\n", "op", "k", "scalar ns/call", "bmi2 ns/call", "speedup"); | ||
|
|
||
| for (int ci=0; ci<2; ci++) { | ||
| int k = counts[ci]; | ||
|
|
||
| // a fixed, strictly-increasing index set and a value mask consistent with it | ||
| int inds[8]; | ||
| for (int i=0; i<k; i++) inds[i] = 3*i + 1; | ||
| uint64_t posMask = makePosMask(inds, k); | ||
| uint64_t valueMask = posMask & 0xA5A5A5A5A5A5A5A5ULL; | ||
|
|
||
| volatile uint64_t sink = 0; | ||
|
|
||
| // ---- gather (getValueOfBits) ---- | ||
| double sg = timeMin(iters, reps, [&](uint64_t N){ | ||
| uint64_t acc = 0; | ||
| for (uint64_t n=0; n<N; n++) acc ^= scalarGather(n, inds, k); | ||
| sink ^= acc; | ||
| }); | ||
| #ifdef BENCH_USE_BMI2 | ||
| double bg = timeMin(iters, reps, [&](uint64_t N){ | ||
| uint64_t acc = 0; | ||
| for (uint64_t n=0; n<N; n++) acc ^= (uint64_t) _pext_u64(n, posMask); | ||
| sink ^= acc; | ||
| }); | ||
| printf("%-8s %-4d %14.3f %14.3f %9.2fx\n", "gather", k, | ||
| nsPerCall(iters, sg), nsPerCall(iters, bg), sg/bg); | ||
| #else | ||
| printf("%-8s %-4d %14.3f %14s %10s\n", "gather", k, nsPerCall(iters, sg), "-", "-"); | ||
| #endif | ||
|
|
||
| // ---- scatter (insertBitsWithMaskedValues) ---- | ||
| double ss = timeMin(iters, reps, [&](uint64_t N){ | ||
| uint64_t acc = 0; | ||
| for (uint64_t n=0; n<N; n++) acc ^= scalarScatter(n, inds, k, valueMask); | ||
| sink ^= acc; | ||
| }); | ||
| #ifdef BENCH_USE_BMI2 | ||
| double bs = timeMin(iters, reps, [&](uint64_t N){ | ||
| uint64_t acc = 0; | ||
| for (uint64_t n=0; n<N; n++) acc ^= (valueMask | (uint64_t) _pdep_u64(n, ~posMask)); | ||
| sink ^= acc; | ||
| }); | ||
| printf("%-8s %-4d %14.3f %14.3f %9.2fx\n", "scatter", k, | ||
| nsPerCall(iters, ss), nsPerCall(iters, bs), ss/bs); | ||
| #else | ||
| printf("%-8s %-4d %14.3f %14s %10s\n", "scatter", k, nsPerCall(iters, ss), "-", "-"); | ||
| #endif | ||
|
|
||
| #ifdef BENCH_USE_BMI2 | ||
| // sanity: the intrinsic and scalar paths must agree (bit-for-bit) for these sorted indices | ||
| bool ok = true; | ||
| for (uint64_t n=0; n<4096 && ok; n++) { | ||
| if ((uint64_t)_pext_u64(n, posMask) != scalarGather(n, inds, k)) ok = false; | ||
| if ((valueMask | (uint64_t)_pdep_u64(n, ~posMask)) != scalarScatter(n, inds, k, valueMask)) ok = false; | ||
| } | ||
| printf(" (k=%d results verified bit-identical to scalar: %s)\n", k, ok ? "yes" : "NO"); | ||
| #endif | ||
| (void) sink; | ||
| } | ||
|
|
||
| return 0; | ||
| } |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
TODO: consider enabling by default when detectedly supported by compiler