gonum.org/v1/gonum@v0.14.0/internal/asm/f64/benchAxpy_test.go (about) 1 // Copyright ©2017 The Gonum Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 package f64_test 6 7 import ( 8 "fmt" 9 "testing" 10 11 . "gonum.org/v1/gonum/internal/asm/f64" 12 ) 13 14 const testLen = 1e5 15 16 var ( 17 a = 2.0 18 x = make([]float64, testLen) 19 y = make([]float64, testLen) 20 z = make([]float64, testLen) 21 ) 22 23 func init() { 24 for n := range x { 25 x[n] = float64(n) 26 y[n] = float64(n) 27 } 28 } 29 30 func BenchmarkAxpyUnitary(t *testing.B) { 31 naiveaxpyu := func(a float64, x, y []float64) { 32 for i, v := range x { 33 y[i] += a * v 34 } 35 } 36 tests := []struct { 37 name string 38 f func(a float64, x, y []float64) 39 }{ 40 {"AxpyUnitary", AxpyUnitary}, 41 {"NaiveAxpyUnitary", naiveaxpyu}, 42 } 43 for _, test := range tests { 44 for _, ln := range []uintptr{1, 3, 10, 30, 1e2, 3e2, 1e3, 3e3, 1e4, 3e4, 1e5} { 45 t.Run(fmt.Sprintf("%s-%d", test.name, ln), func(b *testing.B) { 46 b.SetBytes(int64(64 * ln)) 47 x, y := x[:ln], y[:ln] 48 b.ResetTimer() 49 for i := 0; i < b.N; i++ { 50 test.f(a, x, y) 51 } 52 }) 53 } 54 } 55 } 56 57 func BenchmarkAxpyUnitaryTo(t *testing.B) { 58 naiveaxpyut := func(d []float64, a float64, x, y []float64) { 59 for i, v := range x { 60 d[i] = y[i] + a*v 61 } 62 } 63 tests := []struct { 64 name string 65 f func(z []float64, a float64, x, y []float64) 66 }{ 67 {"AxpyUnitaryTo", AxpyUnitaryTo}, 68 {"NaiveAxpyUnitaryTo", naiveaxpyut}, 69 } 70 for _, test := range tests { 71 for _, ln := range []uintptr{1, 3, 10, 30, 1e2, 3e2, 1e3, 3e3, 1e4, 3e4, 1e5} { 72 t.Run(fmt.Sprintf("%s-%d", test.name, ln), func(b *testing.B) { 73 b.SetBytes(int64(64 * ln)) 74 x, y, z := x[:ln], y[:ln], z[:ln] 75 b.ResetTimer() 76 for i := 0; i < b.N; i++ { 77 test.f(z, a, x, y) 78 } 79 }) 80 } 81 } 82 } 83 84 var incsAxpy = []struct { 85 len uintptr 86 inc []int 87 }{ 88 {1, []int{1}}, 89 {2, []int{1, 2, 4, 10}}, 90 {3, []int{1, 2, 4, 10}}, 91 {4, []int{1, 2, 4, 10}}, 92 {5, []int{1, 2, 4, 10}}, 93 {10, []int{1, 2, 4, 10}}, 94 {500, []int{1, 2, 4, 10}}, 95 {1e3, []int{1, 2, 4, 10}}, 96 {1e4, []int{1, 2, 4, 10, -1, -2, -4, -10}}, 97 } 98 99 func BenchmarkAxpyInc(t *testing.B) { 100 naiveaxpyinc := func(alpha float64, x, y []float64, n, incX, incY, ix, iy uintptr) { 101 for i := 0; i < int(n); i++ { 102 y[iy] += alpha * x[ix] 103 ix += incX 104 iy += incY 105 } 106 } 107 tests := []struct { 108 name string 109 f func(alpha float64, x, y []float64, n, incX, incY, ix, iy uintptr) 110 }{ 111 {"AxpyInc", AxpyInc}, 112 {"NaiveAxpyInc", naiveaxpyinc}, 113 } 114 for _, test := range tests { 115 for _, tt := range incsAxpy { 116 for _, inc := range tt.inc { 117 t.Run(fmt.Sprintf("%s-%d-inc(%d)", test.name, tt.len, inc), func(b *testing.B) { 118 b.SetBytes(int64(64 * tt.len)) 119 var idx, tstInc uintptr = 0, uintptr(inc) 120 if inc < 0 { 121 idx = uintptr((-int(tt.len) + 1) * inc) 122 } 123 for i := 0; i < b.N; i++ { 124 test.f(a, x, y, uintptr(tt.len), tstInc, tstInc, idx, idx) 125 } 126 }) 127 } 128 } 129 } 130 } 131 132 func BenchmarkAxpyIncTo(t *testing.B) { 133 naiveaxpyincto := func(dst []float64, incDst, idst uintptr, alpha float64, x, y []float64, n, incX, incY, ix, iy uintptr) { 134 for i := 0; i < int(n); i++ { 135 dst[idst] = alpha*x[ix] + y[iy] 136 ix += incX 137 iy += incY 138 idst += incDst 139 } 140 } 141 tests := []struct { 142 name string 143 f func(dst []float64, incDst, idst uintptr, alpha float64, x, y []float64, n, incX, incY, ix, iy uintptr) 144 }{ 145 {"AxpyIncTo", AxpyIncTo}, 146 {"NaiveAxpyIncTo", naiveaxpyincto}, 147 } 148 for _, test := range tests { 149 for _, tt := range incsAxpy { 150 for _, inc := range tt.inc { 151 t.Run(fmt.Sprintf("%s-%d-inc(%d)", test.name, tt.len, inc), func(b *testing.B) { 152 b.SetBytes(int64(64 * tt.len)) 153 var idx, tstInc uintptr = 0, uintptr(inc) 154 if inc < 0 { 155 idx = uintptr((-int(tt.len) + 1) * inc) 156 } 157 for i := 0; i < b.N; i++ { 158 test.f(z, tstInc, idx, a, x, y, uintptr(tt.len), 159 tstInc, tstInc, idx, idx) 160 } 161 }) 162 } 163 } 164 } 165 }