clock de 1 GHz significa que o tempo por ciclo de clock é de 1 ns (1/109). Por exemplo, para o benchmark CoreMark [Gal-On and Levy 2012] (100.000 iterações), o desempenho no ARM-32 Cortex-A9 é 32.27 B hoje ante- cipam os resultados dos desvios usando preditores de hardware, que podem exceder a pre- cisão de 90% e trabalhar com qualquer tamanho de pipeline. Eles precisam apenas de um mecanismo para liberar Berkeley, 2015. S. Gal-On and M. Levy. Exploring CoreMark - a benchmark maximizing simplicity and efficacy. The Embedded Microprocessor Benchmark Consortium, 2012. Intel Corporation. Intel 64 and IA-32 Architectures

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

markup (we here use @noinline to prevent the optimizer from trying to be too clever and defeat our benchmark): CHAPTER 34. PERFORMANCE TIPS 393 @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not See also [sind], [sinpi], [sincos], [cis]. source Base.cos – Method. cos(x) Compute cosine of x, where x is in radians. See also [cosd], [cospi], [sincos], [cis]. source Base.Math.sincos – Method

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Compute sine of x, where x is in radians.CHAPTER 44. MATHEMATICS 850 See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Compute sine of x, where x is in radians.CHAPTER 44. MATHEMATICS 850 See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0

(we here use @noinline to prevent the opti- mizer from trying to be too clever and defeat our benchmark): @noinline function inner(x, y) s = zero(eltype(x)) for i=eachindex(x) @inbounds s += x[i]*y[i] – in general, the speedup will be smaller. (In this particular example, the working set of the benchmark is small enough to fit into the L1 cache of the processor, so that memory access latency does not Base.sin – Method. sin(x) Compute sine of x, where x is in radians. See also sind, sinpi, sincos, cis, asin. Examples julia> round.(sin.(range(0, 2pi, length=9)'), digits=3) 1×9 Matrix{Float64}: 0