At some point I have to perform the following loops to visit the colors in a certain radius : for (int. Replace the sigma and radius parameters with literal values to let the Unity shader compiler produce efficient code. // Documentation Portal . Unless you KNOW that you need these and why, you should never need to make use of them. A unity-specific ShaderLab language is used define the shader properties, subshaders and passes, while actual shader code is written in HLSL (High Level Shading Language). Nov 2, 2018HLSL is the native shading language for DirectX, so certain HLSL features do not have corresponding mappings in Vulkan, and certain Vulkan specific information does not. " Code (CSharp): const float minLayers = 8; const float maxLayers = 32; float numLayers = lerp (maxLayers. Reload to refresh your session. But although they should be available the texture object PostProcessInput_0_Texture and sampler object PostProcessInput_0_Sampler defined in PostProcessMaterialShaders. Then we calculate the distance to the input value based on. hlsl. Discussion in 'Shader Graph' started by NarberalTools, May 3, 2021. Also, the shader now appears to compile much faster. Feedback. For example you can write the following (dummy) shader program test. Loop unrolling, also known as loop unwinding, is a loop transformation technique that attempts to optimize a program's execution speed at the expense of its binary size, which is an approach known as space–time tradeoff. flatten. This state includes the runtime mode that the processor is in (e. A MxN HLSL float matrix is translated into a SPIR-V matrix with M vectors, each with N elements. 따라서 hlsl을 배우시면 다른 쉐이더 언어를 익히시는데도 큰 무리가 없을 것입니다. It is available in the latest compiler code, but is not the default. Note This function is part of the HLSL shader linking technology that you can use on all Direct3D 11 platforms to create precompiled HLSL functions, package them into libraries, and link them into full shaders at run time. x が有限の. Touching this variable in shader caused compiler to unroll. jpg","path":"Example. What does it mean? “If” flow controls in HLSL come in two flavors: “branch” and “flatten. Sorted by: 7. . hlsl". 8. . int count = min (iterCount, 4); for (int i = 0; i < count; ++i) {. Sc4Freak. Unroll / loop attributes as seen in other compilers are always seen as hints. See also. fx shader file uses some uniform parameters for a function: VSOutputTxEnvMap ComputeEnvMapVSOutput (VSInputNmTx vin, float3 normal,. The reason it's forced to unroll that loop is because of that texture load and the way D3D handles automatic mip level calculations. Compile with FXC to an fx target pre-compiled binary (ie: fx_2_0 or whatever yoFix unroll, when storing to pointer to array. sponsored post. 0 HDRP (windows) and getting warnings. It is particularly useful in shader model 4 and later when the compiler compiles loops. In the small example shown here, because of an imperfect loop nest. It is executed concurrently on many GPU cores. This type of warning is basically the compiler informing you that a loop can't be unrolled or it would be less performant to do so (as can be read in the Microsoft documentation for the HLSL for-loop). No such reservation for the fmod function as far as I can see. The [branch] and [flatten] tags are just hints to the compiler to prefer one or the other if supported and possible. HLS优化设计中pipeline以及unroll指令：细粒度并行优化的完美循环-HLS 优化设计的最关键指令有两个：一个是流水线 (pipeline) 指令，一个是数据流(dataflow) 指令。正确地使用好这两个指令能够增强算法地并行性，提升吞吐量，降低延迟但是需要遵循一定的代. However it says The % operator is defined only in cases where either both sides are positive or both sides are negative. “Flatten” says “Run both the if and else blocks, but only use the one that is appropriate based. A flow-control statement determines at run time which block of HLSL statements to execute next. Compilation time is with fxc (Windows 8 SDK) about 90 seconds. Use the clip HLSL intrinsic function to simulate clipping planes if each component of the x parameter represents the distance from a plane. This includes code to declare and initialize variables, write. The converter can be downloaded from this page. This could be a problem for us. . . The compiler simulates loops by default to evaluate whether it can unroll them. g. hlsl","path":"Modules/Rendering/Shaders/vaASSAOLite. Although you can use the geometry shader to sub-divide primitives, I would suggest looking into tessellation. シェーダープロパティでピクセル単位で幅を指定できるようにしました。. The OpenGL Shading Language is a C-style language, so it covers most of the features you would expect with such a language. If you comment out the blendedBiome [biome] line you should find it compiles fine. 添加了loop标签的for循环不能展开，流式控制每次. I added that line "compile_arguments -unroll none" as described by Sinbad, but unfortunately that did not have any effect. True if all components of the x parameter are non-zero; otherwise, false. Was this page helpful? Yes No. This function is similar to the any HLSL intrinsic function. Instead of editing intricate code and custom libraries inside. As far as I can tell, that is not true. The second part is a bit more complicated. This function is supported in the following shader models. #3. Thus, I do not need to unroll L0 loop. HLSL中用于优化的指令，[branch],[flatten],[loop],[unroll] branch，shader会根据判断语句只执行当前情况的代码; flatten，shader会执行全部情况的分支代码，然后再根据判断条件获得结果; unroll，for循环是展开的，直到循环条件终止; loop，for循环不展开，Exit the surrounding loop (do, for, while). To review, open the file in an editor that reveals hidden Unicode characters. This topic should be used as a reference point when porting your existing high-level shader language (HLSL) shaders over from D3DCompiler (FXC) to DXCompiler (DXC). Loops in the C/C++ functions are kept rolled by default. On modern processors, loop unrolling. You can now read the entire book for free! GDC Talks. Q&A for work. The hlsl compiler thinks you want your float4s in registers even though your gpu might not have float4 registers at all. Loops can either be unrolled so that there's no loop instructions in the compiled code or left as a proper loop. Fragment Shader. The reason it's forced to unroll that loop is because of that texture load and the way D3D handles automatic mip level calculations. Matteo Bruni (＠Mystral) 14 Apr 2023 14 Apr '23Unity ID. Width, size. . High-level language GLSL & HLSL HLSL Intermediate language SPIR-V SPIR-V & DXIL Supported platform Windows, Linux, macOS Windows, Linux, macOS Compiler library size Small Big (LLVM/Clang) HLSL shader model Up to 5. unroll. Log In to HSL. but I have not got really smart from it yet. Dang man you saved my life! Allthough i was using light cookies i choose the ability to build. soramamenatan. Glossary programs. A pattern such as for (int i = 0; i < n && i < MAX_N; ++i) gives a dynamic bound and allows the compiler to unroll (to nested if-statements). 2. I shall also try the HLSL idea, but that takes me a bit more time to figure out what the differences were between Cg and HLSL again. Loops in the C/C++ functions are kept rolled by default. Where HLSL mostly fails (for AnKi) is in the SPIR-V backend of DXC. It seems HLSL is prohibiting defining a struct nested inside a struct, so make sure to define your custom structs above and outside struct Functions. Sometimes, if you put a compiler-visible upper bound on the loop count then the compiler will be able to unroll it: uniform int iterCount;. In Unity, you use the HLSL programming language to write shader A program that runs on the GPU. Use the [unroll(n)]. しかし粗いポリゴンのモデルにひびを入れる場合は分割数を増やしたいなど、分割具合をもっと柔軟に設定できるようにしたいです。. buy doesn't matter. While enabling it, ( [loop] ), you enable the 'flow' control inside this loop. Geometry Shader. 3. My hlsl shader cannot unroll a loop? 0. Compute Shader. New subject: [PATCH v3 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. buy doesn't matter. hlsl This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. 3. Specifically, this topic provides details about the following: Enable some of the old FXC compilation behaviors that are disabled by default on DXC. In HLSL global variables are considered uniform by default. For example, the following pragma is equivalent to the /Os compiler option:. We need to select an Active Target (Universal, HDRP or Visual Effect, depending on which is installed in the project). An unbounded array in HLSL does match a fixed number set with numDescriptors in the descriptor table, and a fixed size in the HLSL does match an. When unrolling a loop, the iteration count has to be a compile-time constant; without the static on there, the compiler has to assume you might change the value of the parameter at runtime. . Other shading languages. Getting There in Style: Intro to interpolation and Control. Settings. Return Value. // Object Declarations Texture2D g_MeshTexture; // Color texture for mesh SamplerState MeshTextureSampler { Filter =. int count = min (iterCount, 4); for (int i = 0; i < count; ++i) {. Here is the source code. 2. Loop Unrolling (unroll Pragma) The Intel® HLS Compiler supports the unroll pragma for unrolling multiple copies of a loop. The shader assembly now has the ability to jump/branch to a new statement of execution, which is noticeable in the compiled code. Pipeline means to instruct a task to execute in a pipeline, allowing the next execution of the task to begin before the current execution is complete. hatenablog. 生成使用流控制执行循环的每个迭代的代码。 与 [unroll] 属性不兼容。 fastopt: 缩短编译时间，但生成较少主动的优化。 如果使用此属性，编译器将不会展开循环。 此属性仅影响支持 中断 指令的着色器模型目标。aah yes, you're right, webgl is quite limited there. [unroll] and the other HLSL attributes came in a later release of the shader compiler. HLSL shaders can be compiled at author-time or at runtime, and set at runtime into the appropriate pipeline stage. New subject: [PATCH v5 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. GCN3 introduced 16 bit register support. Note the configure may fail if linker plugin. New subject: [PATCH v4 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. It will work on any target that supports the break instruction, though there is a limitation on SM3 pixel shaders, due to complexities involved with gradient oThat sounds great! I'm looking forward to it, and the limitation is reasonable and "avoidable enough" to not really hurt. So you can now do this: template<int iStartPlane, int iEndPlane> bool FrustumCull_Box (float3 BoxOrigin) { [unroll] for (uint PlaneIndex = iStartPlane; PlaneIndex < iEndPlane; ++PlaneIndex) { float4 PlaneData. -Put in blank textures to sample for the samplers that are unused, so, you always look them up. 1, + 6. Sorry. - manually unroll loops (works better (in terms of compilation time) than using [unroll], [fastopt] or whatever compiler hints) - especially true for nested loops! - the deeper the called function, the worse - look for redundant texture sampling which could be pulled up from loops or functions - you'll get cache hit, however it will compile longerShader Model 2 (DirectX HLSL) and higher shader models: yes: Shader Model 1 (DirectX HLSL) vs_1_1: See also. Hi all, There are some [Attribute] blocks in HLSL flow control statements: loop, unroll (x) -> for loop. In your case it's doing this, and since you don't provide an upper bound for "myInt" it doesn't. Vertex Shader. Featured on Meta Update: New Colors Launched. The useful pragmas are 3: #pragma anki include “path/filename. #pragma token-string. com. Ask Question Asked 2 years, 6 months ago. In this article. We start by simply dividing our space into cells by flooring the input value and generating random positions inside of the cells based on that. More info. When unrolling a loop, the iteration count has to be a compile-time constant; without the static on there, the. In general the HLSL compiler likes to unroll loops whenever it can, which is typically when the number of iterations is known at compile time. 特定のサンプル インデックスのサンプル位置 (x,y) を返します。. // the shader code would be a ton more complex. This section of the manual includes information on using HLSL in a Unity-specific way. Core ARB extension. It is a new feature that has been added in DirectX 11, which allows you to sub-divide low level primitives into high-detailed primitives. 3. flatten. Shader Model 4. . 0 and 3. Preprocessor directive that provides machine-specific or operating system-specific features while retaining overall compatibility with the C and C++ languages. 0: 63: October 16, 2023. So i +1 and accept the answer. When I’m trying to. ) basic types: all int/uint/float/double scalar/vector/matrix. A great starting point on using HLSL in Vulkan via SPIR-V is the HLSL to SPIR-V feature mapping manual. abs (ddx (x)) + abs (ddy (x)) を返します。. Shader Model 4. Does the ATI or nVidia driver unroll the loops (SM 2. When using DXC to compile the HLSL produced by spirv-cross I get the following error: error: use of undeclared identifier '_189' ui. hlsl","contentType. "3. Started by 51mon October 18, 2007 10:40 AM. Description. For really basic shaders for texture to texture rendering/blending etc. Minimum Shader Model. One, the Microsoft-style C Attribute syntax, which uses single brackets [] to enclose an attribute and it’s arguments: [WaveOpsIncludeHelperLanes] [shader("compute")] [numthreads(1,1,1)] The second, the HLSL annotation syntax, which annotates variable, field and parameter declarations. generic preprocessor (same as GLSL) scanner (keywords, identifiers, etc. Use flow-control statements in the. Problems building or running the engine, queries about how to use features etc. from bool. The ShaderLab syntax hasn’t changed much compared to the built-in pipeline. Defines the number of threads to be executed in a single thread group when a compute shader is dispatched (see ID3D11DeviceContext::Dispatch ). you are not allowed to have dynamic variables in the 'for'. Flow Control See moreJan 4, 2013. HLSL compilation speed. Low fps on Arm Mali GPU. Minimal working example Below is an MWE consisting of a C# script for Unity, the corresponding compute shader where the computations are performed and a screen-shot of my console when. A Compute Shader is a Shader Stage that is used entirely for computing arbitrary information. // vertex #version 450 #extension GL_ARB_separate_shader_objects : enable out gl_PerVertex. @user2464424: Yes, NV does have quite a few proprietary GLSL #pragma directives. . New subject: [PATCH v5 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. I believe so, yes. Submit and view feedback for. Rendering. The Xilinx Vitis-HLS synthesises the for -loop into a pipelined microarchitecture with II=1. To compute a smooth interpolation between 0 and 1, use the smoothstep HLSL intrinsic. Re: [PATCH v8 0/1] MR150: vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. If that's what's going on, then Naga may just have to accept. Unroll means that the same sort of function which should be done n-times should be done in (sem-)parallel to get a shorter latency and higher thoughput. The decoder ring also has a translation table for concepts and terms used in. ArachnidAnimal, Nov 25, 2017. ). January 19, 2009 10:59 PM. That is, if we can assume it's the same situation for PC. This is the home page of the book 3D Math Primer for Graphics and Game Development. It takes an array like { 1, 2, 1, 2 } and outputs the accumulative sum from left to right either exclusive { 0, 1, 3, 4 } or inclusive { 1, 3, 4, 6 }. When selected no message appears. . g. A flow-control statement determines at run time which block of HLSL statements to execute next. New Here? Create an AccountHLSL is the native shading language for DirectX, so certain HLSL features do not have corresponding mappings in Vulkan, and certain Vulkan specific information does not have native ways to express in HLSL source code. 설명 [unroll] 및 [loop] 특성은 상호 배타적이며 둘 다 지정된 경우 컴파일러 오류를 생성합니다. We just do a best effort for backcompat with FXC because it would accept some surprising patterns, but it is impossible for us to match behavior in all cases. For our implementation of voronoi noise each of our cells will have one point. The regular for loops in HLSL compute shaders gives different results depending on the local computer it is ran on (see image below). Resources Developer Site; Xilinx Wiki; Xilinx GithubNew subject: [PATCH v4 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. See also. In HLSL, #pragma directives are a type of preprocessor directive. hlsl. Nikolay Sivov 14 Apr 2023 14 Apr '23Unfortantly, especially for pixel shader targets, there are very few cases where you can actually use a loop to do anything useful, most of the time the time HLSL has to unroll it. Problems building or running the engine, queries about how to use features etc. The UNROLL pragma transforms loops by creating multiples copies of the loop body in the RTL design, which allows some or all loop iterations to occur in parallel. I notice that if I allocate a Pool. You need to declare blurLength as a static int for this to work. 0), Metal (MSL) and also other GLSL versions (GLSL 330, GLSL 400, etc. The #pragma unroll directive however can be used to control unrolling. // Object Declarations Texture2D g_MeshTexture; // Color texture for mesh SamplerState MeshTextureSampler { Filter =. One way is using the HLS pragma as follows: const unsigned int N = 1024; const unsigned int B =. Each line begins with a regular expression followed by a statement written in C code. I think that would require manually doing. hlsl -spirv -enable-16bit-types says dxc failed : enable-16bit-types is only allowed for shader model >= 6. {"payload":{"allShortcutsEnabled":false,"fileTree":{"OIT_DX11/AOIT Technique":{"items":[{"name":"AOIT. Typically, pixels outnumber vertices in a scene, so pixel shaders execute more often than vertex shaders. I don't want the loop to be unrolled, so I placed a [loop] attribute above it. to name a few examples: the ternary operator (condition ? a : b), range-based for syntax (for (variable : collection)), and switch label marking (case 1:). [unroll] turns into spv::LoopControlUnrollMask [loop] turns in. The following code shows the HLSL closest hit shader that calculates shadows with a single point light from the sample app: #include "raycommon. This state includes the runtime mode that the processor is in (e. Image Effect Shaders have culling off by default if you’d like to see an example. I've tried various different things to try and fix such as using the [loop] attribute, in which case I get “can't unroll loops marked with loop attribute”, and I've tried using the [unroll()] attribute to specify the amount of iterations the loop goes through, but I get the unrolling failed message again. The above code forces the loop to unroll and gives the following warning: warning X3550: array reference cannot be used as an l-value; not natively addressable, forcing loop to unroll. Reduces the compile time but produces less aggressive optimizations. Connect and share knowledge within a single location that is structured and easy to search. I’ve successfully build all the binaries and trying to open the starter content. The decoder ring also has a translation table for concepts and terms used in. 3. Might be cheaper than an if, might not. Or in cases where both sides should be shown, you can turn culling off. When loops are rolled, synthesis creates the logic for one iteration of the loop, and the RTL design executes this logic for. So I was trying to work this around and use a non-gradient sample method from hlsl documentation called SampleLevel. 1 is available in Direct3D 10. フラグメントシェーダーにて重心座標系の値の偏微分を取り、1ピクセルでの変化量を求めます。. compute(line_num) (on d3d11) I've had a look through the Microsoft docs and some other forums, but I don't understand many of the solutions and most of the answers are for vertex shaders and. (UNITY_COMPILER_HLSL) [unroll(16)] #endif Or just remove that and let the shader. First, the reason why the second one works, is because it is automatically unrolled. In HLSL, #include directives are a type of preprocessor directive. It contains detailed information on semantics, syntax, supported features and extensions and much more and is a must-read. With these requirements, I put the following constraints: #pragma HLS LATENCY min=500 max=528 // directive for FUNCT #pragma HLS UNROLL factor=1 // directive for L0 loop However, the synthesized design results in function latency over 3000 cycles and the log shows the following warning. Parameters. void main () { return ; } However, a return statement can return one or more values. txt file in the build directory. They instruct the compiler to include the contents of one HLSL file inside another. gerstner_wave. The source file is hlsl_lexer. Unrolling is generally. Unroll the loop until it stops executing. numthreads. Anyway the HLSL compiler will, in general, attempt to unroll loops and flatten branches unless you try to convince it otherwise. In conclusion . HLSL is the C-like high-level shader language that you use with programmable shaders in DirectX. I think it is because after compilation the shader may try to run. _BaseMap、_BumpMap、_EmissionMap定义; 空间变换 (Space Transform) SpaceTransforms. The text was updated successfully, but these errors were encountered:For the SPIR-V example contained in this archive, spirv-cross seems to be generating incorrect HLSL code. Hello, I was looking for a simple way to do edge highlighting and I could not find any solid configurable options. New subject: [PATCH v8 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. bool invertSource = false, invertDepth = false; float2 sourceUVC = inputUV, depthUVC = inputUV; // these need to be in the function itself; if they were consts outside then they would be part of the CBUFFER and. hlsl This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. The loop condition depends on shader input, so I don't think that the processor is detecting a loop that it. branch, flatten -> if statement. Stack Exchange network consists of 181 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers. When unrolling a loop, the iteration count has to be a compile-time constant; without the static on there, the compiler has to assume you might change the value of the parameter at runtime. 1), when I navigate to the URP Lit shader in my project directory (Packages > Universal RP > Shaders > Lit) and select it, I see that it has 2 errors, both about not being able to open include file 'LitInput. To review, open the file in an editor that reveals hidden Unicode characters. Here is shader code: void DownsamplePass_Pixel_Shader ( in float4 Position : POSITION0, in float4 PositionSS :. 1 containing targeted fixes for issues uncovered by Wine tests, in particular an assert which is currently preventing Wine tests from running with 1. [Speculation mode] The reason behind this is that perhaps some hardware systems do not have true array indexing or looping and so the compiler simply unwind the loops. I have to do the Filtering with the variable filter size. Unroll and loop work alongside the optimizing compiler, but are not substitute for it, nor can they be substituted by it. for语句. 1 or higher. New subject: [PATCH v5 1/1] vkd3d-shader/hlsl: Ignore "unroll" attribute for loops. What you get depends on the kind of conversion. Loop Unrolling (unroll Pragma) The Intel® HLS Compiler supports the unroll pragma for unrolling multiple copies of a loop. HLSL semantics. Append(o) stream. g Sample in non-uniform control flow) and try to unroll the loop. Though obviously the shader would no longer do the job you want. Sometimes, if you put a compiler-visible upper bound on the loop count then the compiler will be able to unroll it: uniform int iterCount;. Supported. The regular for loops. An optional parameter that controls how the statement is compiled. hlsl","contentType. Can somebody help me with it? Update: I think I found HLSL code with that errors but still don’t know how to fix it:1 Answer. Adding the unroll attribute to loops does not guarantee that the compiler is able to legalize the code. In short: You can't return an array of floats defined in the function in HLSL. More info See in Glossary compiler that isn’t covered by other types of preprocessor directive. shaders-msl-no-opt. Posts: 198. TextureCubeArray is available in Shader Model 4. // vertex #version 450 #extension GL_ARB_separate_shader_objects : enable out gl_PerVertex. \$\begingroup\$ @DMGregory, actually just being able to do it in the Unity editor would be fine if it is possible to create a compute shader asset through a script. This could be a reason that fxc try to unroll the for loop so that the shader program is executed in lockstep mode. Language Syntax (DirectX HLSL) - Programming shaders in HLSL requires that you understand the language syntax, that is, how you write HLSL code. . Optionally, you can specify the maximum number of times the loop can execute. Array reference can not be used as an l-value. The HLSL definition means that using the [unroll(count)] can cut off loop iterations before the loop condition would have stopped iteration, and the loop is guaranteed to be completely unrolled. In the next compiler release, we've added an attribute called [fastopt] that tells the compiler not to bother simulating the loop. ShaderLab. Type Description. Minimal working example Below is an MWE consisting of a C# script for Unity, the corresponding compute shader where the computations are performed and a screen-shot of my console when. 相关主题. You signed out in another tab or window. It's usually better to unroll the stuff yourself, since AMD/Intel/. I assume this can be safely accepted. I couldn't find any direct proof of this, but I think you might find that you can only access the "array" using the iterated variable (i. Play around with sigma and radius values to find the optimal parameters. But they're no-ops on anything but HLSL. Use '-HV 2016' to treat this as warning. 2. Instead I’m going to throw my Material asset (including comments) at you for everyone to use, adjust and play around with. void main () { return ; } However, a return statement can return one or more values. We would like to show you a description here but the site won’t allow us. Use the [loop] attribute in your HLSL to force actual loops. Though obviously the shader would no longer do the job you want. Connect and share knowledge within a single location that is structured and easy to search. 5) ? <complex formula> : 0;You signed in with another tab or window. September 28, 2023 23:58. 1 I didn't pick any of them. To add HLSL code to your shader asset, you put the code inside a shader code block. Creates a linker interface. Many of them are very similar or identical to C++ counterparts. First of all, I created a custom node which allows me to perform a Gaussian Blur on a Texture Object. The HLSL parser is built in two parts: the lexer and the parser. There's probably crap I'm missing but GLSL needs some help. OpenGL has vender specific unrolling and stuff. The following information only applies to. These parameters are the same letters used with the /O compiler options. "3. That's a no no. See also. January 27, 2011 06:25 AM. As far as I know, you have to declare arrays with a predefined length ( Array syntax needs a positive integer as size ). I would like for this code to be dynamic so as to reduce the instruction count. This is a design decision. Add test for sample mask bulk load/store. The parser works with preprocessor pragmas and the syntax is similar to c++ OpenMP. Remarks. Late response to this, but all I can suggest is leveraging the per-component math of HLSL better in your shaders. Creates a linker interface. The solution to this issue is either to make sure that the proper side is always facing the camera. On modern processors, loop unrolling.