[libc][gpu] Add Sinf Benchmarks (#102532)

This PR adds benchmarking for `sinf()` using the same set up as `sin()`
but with a smaller range for floats.

Author: jameshu15869

libc/benchmarks/gpu/src/math/CMakeLists.txt

@@ -32,6 +32,7 @@ add_benchmark(
     sin_benchmark.cpp
   DEPENDS
     libc.src.math.sin
+    libc.src.math.sinf
     libc.src.stdlib.srand
     libc.src.stdlib.rand
     libc.src.__support.FPUtil.fp_bits

libc/benchmarks/gpu/src/math/sin_benchmark.cpp

@@ -5,6 +5,7 @@
 #include "src/__support/CPP/functional.h"
 #include "src/__support/FPUtil/FPBits.h"
 #include "src/math/sin.h"
+#include "src/math/sinf.h"
 #include "src/stdlib/rand.h"
 
 #ifdef NVPTX_MATH_FOUND
@@ -19,37 +20,56 @@
 // uint64_t representing the latency. Defining each benchmark using macro that
 // expands to a lambda to allow us to switch the implementation of `sin()` to
 // easily register NVPTX benchmarks.
-#define BM_RANDOM_INPUT(Func, MIN_EXP, MAX_EXP, N)                             \
+#define BM_RANDOM_INPUT(T, Func, MIN_EXP, MAX_EXP, N)                          \
   []() {                                                                       \
-    return LIBC_NAMESPACE::benchmarks::MathPerf<                               \
-        double>::run_throughput_in_range<N>(Func, MIN_EXP, MAX_EXP);           \
+    return LIBC_NAMESPACE::benchmarks::MathPerf<T>::run_throughput_in_range<   \
+        N>(Func, MIN_EXP, MAX_EXP);                                            \
   }
 
-#define BENCH(Name, Func, MIN_EXP, MAX_EXP)                                    \
+#define BENCH(T, Name, Func, MIN_EXP, MAX_EXP)                                 \
   SINGLE_WAVE_BENCHMARK(LlvmLibcSinGpuBenchmark, Name##_1,                     \
-                        BM_RANDOM_INPUT(Func, MIN_EXP, MAX_EXP, 1));           \
+                        BM_RANDOM_INPUT(T, Func, MIN_EXP, MAX_EXP, 1));        \
   SINGLE_WAVE_BENCHMARK(LlvmLibcSinGpuBenchmark, Name##_128,                   \
-                        BM_RANDOM_INPUT(Func, MIN_EXP, MAX_EXP, 128));         \
+                        BM_RANDOM_INPUT(T, Func, MIN_EXP, MAX_EXP, 128));      \
   SINGLE_WAVE_BENCHMARK(LlvmLibcSinGpuBenchmark, Name##_1024,                  \
-                        BM_RANDOM_INPUT(Func, MIN_EXP, MAX_EXP, 1024));        \
+                        BM_RANDOM_INPUT(T, Func, MIN_EXP, MAX_EXP, 1024));     \
   SINGLE_WAVE_BENCHMARK(LlvmLibcSinGpuBenchmark, Name##_4096,                  \
-                        BM_RANDOM_INPUT(Func, MIN_EXP, MAX_EXP, 4096))
+                        BM_RANDOM_INPUT(T, Func, MIN_EXP, MAX_EXP, 4096))
 
-BENCH(Sin, LIBC_NAMESPACE::sin, -1023, 1023);
-BENCH(SinTwoPi, LIBC_NAMESPACE::sin, -10, 3);
-BENCH(SinTwoPow30, LIBC_NAMESPACE::sin, 0, 30);
-BENCH(SinVeryLarge, LIBC_NAMESPACE::sin, 30, 1000);
+BENCH(double, Sin, LIBC_NAMESPACE::sin, -1023, 1023);
+BENCH(double, SinTwoPi, LIBC_NAMESPACE::sin, -10, 3);
+BENCH(double, SinTwoPow30, LIBC_NAMESPACE::sin, 0, 30);
+BENCH(double, SinVeryLarge, LIBC_NAMESPACE::sin, 30, 1000);
 
 #ifdef NVPTX_MATH_FOUND
-BENCH(NvSin, LIBC_NAMESPACE::__nv_sin, -1023, 1023);
-BENCH(NvSinTwoPi, LIBC_NAMESPACE::__nv_sin, -10, 3);
-BENCH(NvSinTwoPow30, LIBC_NAMESPACE::__nv_sin, 0, 30);
-BENCH(NvSinVeryLarge, LIBC_NAMESPACE::__nv_sin, 30, 1000);
+BENCH(double, NvSin, LIBC_NAMESPACE::__nv_sin, -1023, 1023);
+BENCH(double, NvSinTwoPi, LIBC_NAMESPACE::__nv_sin, -10, 3);
+BENCH(double, NvSinTwoPow30, LIBC_NAMESPACE::__nv_sin, 0, 30);
+BENCH(double, NvSinVeryLarge, LIBC_NAMESPACE::__nv_sin, 30, 1000);
 #endif
 
 #ifdef AMDGPU_MATH_FOUND
-BENCH(AmdSin, LIBC_NAMESPACE::__ocml_sin_f64, -1023, 1023);
-BENCH(AmdSinTwoPi, LIBC_NAMESPACE::__ocml_sin_f64, -10, 3);
-BENCH(AmdSinTwoPow30, LIBC_NAMESPACE::__ocml_sin_f64, 0, 30);
-BENCH(AmdSinVeryLarge, LIBC_NAMESPACE::__ocml_sin_f64, 30, 1000);
+BENCH(double, AmdSin, LIBC_NAMESPACE::__ocml_sin_f64, -1023, 1023);
+BENCH(double, AmdSinTwoPi, LIBC_NAMESPACE::__ocml_sin_f64, -10, 3);
+BENCH(double, AmdSinTwoPow30, LIBC_NAMESPACE::__ocml_sin_f64, 0, 30);
+BENCH(double, AmdSinVeryLarge, LIBC_NAMESPACE::__ocml_sin_f64, 30, 1000);
+#endif
+
+BENCH(float, Sinf, LIBC_NAMESPACE::sinf, -127, 128);
+BENCH(float, SinfTwoPi, LIBC_NAMESPACE::sinf, -10, 3);
+BENCH(float, SinfTwoPow30, LIBC_NAMESPACE::sinf, 0, 30);
+BENCH(float, SinfVeryLarge, LIBC_NAMESPACE::sinf, 30, 120);
+
+#ifdef NVPTX_MATH_FOUND
+BENCH(float, NvSinf, LIBC_NAMESPACE::__nv_sinf, -127, 128);
+BENCH(float, NvSinfTwoPi, LIBC_NAMESPACE::__nv_sinf, -10, 3);
+BENCH(float, NvSinfTwoPow30, LIBC_NAMESPACE::__nv_sinf, 0, 30);
+BENCH(float, NvSinfVeryLarge, LIBC_NAMESPACE::__nv_sinf, 30, 120);
+#endif
+
+#ifdef AMDGPU_MATH_FOUND
+BENCH(float, AmdSinf, LIBC_NAMESPACE::__ocml_sin_f32, -127, 128);
+BENCH(float, AmdSinfTwoPi, LIBC_NAMESPACE::__ocml_sin_f32, -10, 3);
+BENCH(float, AmdSinfTwoPow30, LIBC_NAMESPACE::__ocml_sin_f32, 0, 30);
+BENCH(float, AmdSinfVeryLarge, LIBC_NAMESPACE::__ocml_sin_f32, 30, 120);
 #endif