mirror of
https://github.com/ml-explore/mlx.git
synced 2025-09-20 12:08:14 +08:00
rebase
This commit is contained in:
CircleCI Docs
@@ -130,16 +130,16 @@ $(function(){ initResizable(false); });
|
||||
<div class="line"><a id="l00032" name="l00032"></a><span class="lineno"> 32</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>& b_strides,</div>
|
||||
<div class="line"><a id="l00033" name="l00033"></a><span class="lineno"> 33</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>& c_strides,</div>
|
||||
<div class="line"><a id="l00034" name="l00034"></a><span class="lineno"> 34</span> uint index [[thread_position_in_grid]]) {</div>
|
||||
<div class="line"><a id="l00035" name="l00035"></a><span class="lineno"> 35</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a196a07022b812b241d4c06192c0fa83d">elem_to_loc_1</a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00036" name="l00036"></a><span class="lineno"> 36</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a196a07022b812b241d4c06192c0fa83d">elem_to_loc_1</a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00037" name="l00037"></a><span class="lineno"> 37</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a196a07022b812b241d4c06192c0fa83d">elem_to_loc_1</a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00035" name="l00035"></a><span class="lineno"> 35</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ac612d0ae30b8257198339debe04916a3">elem_to_loc_1<size_t, uint></a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00036" name="l00036"></a><span class="lineno"> 36</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ac612d0ae30b8257198339debe04916a3">elem_to_loc_1<size_t, uint></a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00037" name="l00037"></a><span class="lineno"> 37</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ac612d0ae30b8257198339debe04916a3">elem_to_loc_1<size_t, uint></a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00038" name="l00038"></a><span class="lineno"> 38</span> d[index] = Op()(a[a_idx], b[b_idx], c[c_idx]);</div>
|
||||
<div class="line"><a id="l00039" name="l00039"></a><span class="lineno"> 39</span>}</div>
|
||||
</div>
|
||||
<div class="line"><a id="l00040" name="l00040"></a><span class="lineno"> 40</span> </div>
|
||||
<div class="line"><a id="l00041" name="l00041"></a><span class="lineno"> 41</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op></div>
|
||||
<div class="line"><a id="l00041" name="l00041"></a><span class="lineno"> 41</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op, <span class="keyword">typename</span> IdxT = <span class="keywordtype">size_t</span>></div>
|
||||
<div class="foldopen" id="foldopen00042" data-start="{" data-end="}">
|
||||
<div class="line"><a id="l00042" name="l00042"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#afdf0d9d0cb21fcb3f176500785076af8"> 42</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#afdf0d9d0cb21fcb3f176500785076af8">ternary_g_nd2</a>(</div>
|
||||
<div class="line"><a id="l00042" name="l00042"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#adec9ca8a8bf527cb15d70da5857af15d"> 42</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#adec9ca8a8bf527cb15d70da5857af15d">ternary_g_nd2</a>(</div>
|
||||
<div class="line"><a id="l00043" name="l00043"></a><span class="lineno"> 43</span> device <span class="keyword">const</span> <span class="keywordtype">bool</span>* a,</div>
|
||||
<div class="line"><a id="l00044" name="l00044"></a><span class="lineno"> 44</span> device <span class="keyword">const</span> T* b,</div>
|
||||
<div class="line"><a id="l00045" name="l00045"></a><span class="lineno"> 45</span> device <span class="keyword">const</span> T* c,</div>
|
||||
@@ -149,17 +149,17 @@ $(function(){ initResizable(false); });
|
||||
<div class="line"><a id="l00049" name="l00049"></a><span class="lineno"> 49</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span> c_strides[2],</div>
|
||||
<div class="line"><a id="l00050" name="l00050"></a><span class="lineno"> 50</span> uint2 index [[thread_position_in_grid]],</div>
|
||||
<div class="line"><a id="l00051" name="l00051"></a><span class="lineno"> 51</span> uint2 grid_dim [[threads_per_grid]]) {</div>
|
||||
<div class="line"><a id="l00052" name="l00052"></a><span class="lineno"> 52</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ad6c45cacca97899cd362df49c06fea79">elem_to_loc_2</a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00053" name="l00053"></a><span class="lineno"> 53</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ad6c45cacca97899cd362df49c06fea79">elem_to_loc_2</a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00054" name="l00054"></a><span class="lineno"> 54</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#ad6c45cacca97899cd362df49c06fea79">elem_to_loc_2</a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00055" name="l00055"></a><span class="lineno"> 55</span> <span class="keywordtype">size_t</span> out_idx = index.x + size_t(grid_dim.x) * index.y;</div>
|
||||
<div class="line"><a id="l00052" name="l00052"></a><span class="lineno"> 52</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a43f33efc000962d6de881a3aab7458de">elem_to_loc_2<size_t, IdxT></a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00053" name="l00053"></a><span class="lineno"> 53</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a43f33efc000962d6de881a3aab7458de">elem_to_loc_2<size_t, IdxT></a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00054" name="l00054"></a><span class="lineno"> 54</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a43f33efc000962d6de881a3aab7458de">elem_to_loc_2<size_t, IdxT></a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00055" name="l00055"></a><span class="lineno"> 55</span> IdxT out_idx = index.x + IdxT(grid_dim.x) * index.y;</div>
|
||||
<div class="line"><a id="l00056" name="l00056"></a><span class="lineno"> 56</span> d[out_idx] = Op()(a[a_idx], b[b_idx], c[c_idx]);</div>
|
||||
<div class="line"><a id="l00057" name="l00057"></a><span class="lineno"> 57</span>}</div>
|
||||
</div>
|
||||
<div class="line"><a id="l00058" name="l00058"></a><span class="lineno"> 58</span> </div>
|
||||
<div class="line"><a id="l00059" name="l00059"></a><span class="lineno"> 59</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op></div>
|
||||
<div class="line"><a id="l00059" name="l00059"></a><span class="lineno"> 59</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op, <span class="keyword">typename</span> IdxT = <span class="keywordtype">size_t</span>></div>
|
||||
<div class="foldopen" id="foldopen00060" data-start="{" data-end="}">
|
||||
<div class="line"><a id="l00060" name="l00060"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#a113df0c8a841b0e986900d580644e047"> 60</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#a113df0c8a841b0e986900d580644e047">ternary_g_nd3</a>(</div>
|
||||
<div class="line"><a id="l00060" name="l00060"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#a046dcbf67cd2318d45355dc7516e3ff4"> 60</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#a046dcbf67cd2318d45355dc7516e3ff4">ternary_g_nd3</a>(</div>
|
||||
<div class="line"><a id="l00061" name="l00061"></a><span class="lineno"> 61</span> device <span class="keyword">const</span> <span class="keywordtype">bool</span>* a,</div>
|
||||
<div class="line"><a id="l00062" name="l00062"></a><span class="lineno"> 62</span> device <span class="keyword">const</span> T* b,</div>
|
||||
<div class="line"><a id="l00063" name="l00063"></a><span class="lineno"> 63</span> device <span class="keyword">const</span> T* c,</div>
|
||||
@@ -169,60 +169,58 @@ $(function(){ initResizable(false); });
|
||||
<div class="line"><a id="l00067" name="l00067"></a><span class="lineno"> 67</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span> c_strides[3],</div>
|
||||
<div class="line"><a id="l00068" name="l00068"></a><span class="lineno"> 68</span> uint3 index [[thread_position_in_grid]],</div>
|
||||
<div class="line"><a id="l00069" name="l00069"></a><span class="lineno"> 69</span> uint3 grid_dim [[threads_per_grid]]) {</div>
|
||||
<div class="line"><a id="l00070" name="l00070"></a><span class="lineno"> 70</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a2c34ed54714c69e6e1b44344f9e6e330">elem_to_loc_3</a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00071" name="l00071"></a><span class="lineno"> 71</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a2c34ed54714c69e6e1b44344f9e6e330">elem_to_loc_3</a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00072" name="l00072"></a><span class="lineno"> 72</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a2c34ed54714c69e6e1b44344f9e6e330">elem_to_loc_3</a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00073" name="l00073"></a><span class="lineno"> 73</span> <span class="keywordtype">size_t</span> out_idx =</div>
|
||||
<div class="line"><a id="l00074" name="l00074"></a><span class="lineno"> 74</span> index.x + grid_dim.x * (index.y + size_t(grid_dim.y) * index.z);</div>
|
||||
<div class="line"><a id="l00075" name="l00075"></a><span class="lineno"> 75</span> d[out_idx] = Op()(a[a_idx], b[b_idx], c[c_idx]);</div>
|
||||
<div class="line"><a id="l00076" name="l00076"></a><span class="lineno"> 76</span>}</div>
|
||||
<div class="line"><a id="l00070" name="l00070"></a><span class="lineno"> 70</span> <span class="keyword">auto</span> a_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a650f8ea8cf9f9519da9e301aad0308dc">elem_to_loc_3<size_t, IdxT></a>(index, a_strides);</div>
|
||||
<div class="line"><a id="l00071" name="l00071"></a><span class="lineno"> 71</span> <span class="keyword">auto</span> b_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a650f8ea8cf9f9519da9e301aad0308dc">elem_to_loc_3<size_t, IdxT></a>(index, b_strides);</div>
|
||||
<div class="line"><a id="l00072" name="l00072"></a><span class="lineno"> 72</span> <span class="keyword">auto</span> c_idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a650f8ea8cf9f9519da9e301aad0308dc">elem_to_loc_3<size_t, IdxT></a>(index, c_strides);</div>
|
||||
<div class="line"><a id="l00073" name="l00073"></a><span class="lineno"> 73</span> IdxT out_idx = index.x + grid_dim.x * (index.y + IdxT(grid_dim.y) * index.z);</div>
|
||||
<div class="line"><a id="l00074" name="l00074"></a><span class="lineno"> 74</span> d[out_idx] = Op()(a[a_idx], b[b_idx], c[c_idx]);</div>
|
||||
<div class="line"><a id="l00075" name="l00075"></a><span class="lineno"> 75</span>}</div>
|
||||
</div>
|
||||
<div class="line"><a id="l00077" name="l00077"></a><span class="lineno"> 77</span> </div>
|
||||
<div class="line"><a id="l00078" name="l00078"></a><span class="lineno"> 78</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op, <span class="keywordtype">int</span> N = 1></div>
|
||||
<div class="foldopen" id="foldopen00079" data-start="{" data-end="}">
|
||||
<div class="line"><a id="l00079" name="l00079"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#adf8b5989de971e43829875dc0097cdfb"> 79</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#adf8b5989de971e43829875dc0097cdfb">ternary_g</a>(</div>
|
||||
<div class="line"><a id="l00080" name="l00080"></a><span class="lineno"> 80</span> device <span class="keyword">const</span> <span class="keywordtype">bool</span>* a,</div>
|
||||
<div class="line"><a id="l00081" name="l00081"></a><span class="lineno"> 81</span> device <span class="keyword">const</span> T* b,</div>
|
||||
<div class="line"><a id="l00082" name="l00082"></a><span class="lineno"> 82</span> device <span class="keyword">const</span> T* c,</div>
|
||||
<div class="line"><a id="l00083" name="l00083"></a><span class="lineno"> 83</span> device T* d,</div>
|
||||
<div class="line"><a id="l00084" name="l00084"></a><span class="lineno"> 84</span> constant <span class="keyword">const</span> <span class="keywordtype">int</span>* shape,</div>
|
||||
<div class="line"><a id="l00085" name="l00085"></a><span class="lineno"> 85</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* a_strides,</div>
|
||||
<div class="line"><a id="l00086" name="l00086"></a><span class="lineno"> 86</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* b_strides,</div>
|
||||
<div class="line"><a id="l00087" name="l00087"></a><span class="lineno"> 87</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* c_strides,</div>
|
||||
<div class="line"><a id="l00088" name="l00088"></a><span class="lineno"> 88</span> constant <span class="keyword">const</span> <span class="keywordtype">int</span>& ndim,</div>
|
||||
<div class="line"><a id="l00089" name="l00089"></a><span class="lineno"> 89</span> uint3 index [[thread_position_in_grid]],</div>
|
||||
<div class="line"><a id="l00090" name="l00090"></a><span class="lineno"> 90</span> uint3 grid_dim [[threads_per_grid]]) {</div>
|
||||
<div class="line"><a id="l00091" name="l00091"></a><span class="lineno"> 91</span> <span class="keyword">auto</span> idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a66940b1cc3d64651d24634bc696d528b">elem_to_loc_3_nd</a>(</div>
|
||||
<div class="line"><a id="l00092" name="l00092"></a><span class="lineno"> 92</span> {N * index.x, index.y, index.z},</div>
|
||||
<div class="line"><a id="l00093" name="l00093"></a><span class="lineno"> 93</span> shape,</div>
|
||||
<div class="line"><a id="l00094" name="l00094"></a><span class="lineno"> 94</span> a_strides,</div>
|
||||
<div class="line"><a id="l00095" name="l00095"></a><span class="lineno"> 95</span> b_strides,</div>
|
||||
<div class="line"><a id="l00096" name="l00096"></a><span class="lineno"> 96</span> c_strides,</div>
|
||||
<div class="line"><a id="l00097" name="l00097"></a><span class="lineno"> 97</span> ndim);</div>
|
||||
<div class="line"><a id="l00098" name="l00098"></a><span class="lineno"> 98</span> <span class="keyword">auto</span> xshape = shape[ndim - 1];</div>
|
||||
<div class="line"><a id="l00099" name="l00099"></a><span class="lineno"> 99</span> <span class="keywordtype">size_t</span> out_idx =</div>
|
||||
<div class="line"><a id="l00100" name="l00100"></a><span class="lineno"> 100</span> N * index.x + xshape * (index.y + size_t(grid_dim.y) * index.z);</div>
|
||||
<div class="line"><a id="l00101" name="l00101"></a><span class="lineno"> 101</span> <span class="keyword">auto</span> a_xstride = a_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00102" name="l00102"></a><span class="lineno"> 102</span> <span class="keyword">auto</span> b_xstride = b_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00103" name="l00103"></a><span class="lineno"> 103</span> <span class="keyword">auto</span> c_xstride = c_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00104" name="l00104"></a><span class="lineno"> 104</span> <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i < N && (int(N * index.x) + i) < xshape; ++i) {</div>
|
||||
<div class="line"><a id="l00105" name="l00105"></a><span class="lineno"> 105</span> d[out_idx++] = Op()(a[idx.x], b[idx.y], c[idx.z]);</div>
|
||||
<div class="line"><a id="l00106" name="l00106"></a><span class="lineno"> 106</span> idx.x += a_xstride;</div>
|
||||
<div class="line"><a id="l00107" name="l00107"></a><span class="lineno"> 107</span> idx.y += b_xstride;</div>
|
||||
<div class="line"><a id="l00108" name="l00108"></a><span class="lineno"> 108</span> idx.z += c_xstride;</div>
|
||||
<div class="line"><a id="l00109" name="l00109"></a><span class="lineno"> 109</span> }</div>
|
||||
<div class="line"><a id="l00110" name="l00110"></a><span class="lineno"> 110</span>}</div>
|
||||
<div class="line"><a id="l00076" name="l00076"></a><span class="lineno"> 76</span> </div>
|
||||
<div class="line"><a id="l00077" name="l00077"></a><span class="lineno"> 77</span><span class="keyword">template</span> <<span class="keyword">typename</span> T, <span class="keyword">typename</span> Op, <span class="keywordtype">int</span> N = 1, <span class="keyword">typename</span> IdxT = <span class="keywordtype">size_t</span>></div>
|
||||
<div class="foldopen" id="foldopen00078" data-start="{" data-end="}">
|
||||
<div class="line"><a id="l00078" name="l00078"></a><span class="lineno"><a class="line" href="metal_2kernels_2ternary_8h.html#ab2051fd944c2e24c57d5b4af54894d72"> 78</a></span>[[kernel]] <span class="keywordtype">void</span> <a class="code hl_function" href="metal_2kernels_2ternary_8h.html#ab2051fd944c2e24c57d5b4af54894d72">ternary_g</a>(</div>
|
||||
<div class="line"><a id="l00079" name="l00079"></a><span class="lineno"> 79</span> device <span class="keyword">const</span> <span class="keywordtype">bool</span>* a,</div>
|
||||
<div class="line"><a id="l00080" name="l00080"></a><span class="lineno"> 80</span> device <span class="keyword">const</span> T* b,</div>
|
||||
<div class="line"><a id="l00081" name="l00081"></a><span class="lineno"> 81</span> device <span class="keyword">const</span> T* c,</div>
|
||||
<div class="line"><a id="l00082" name="l00082"></a><span class="lineno"> 82</span> device T* d,</div>
|
||||
<div class="line"><a id="l00083" name="l00083"></a><span class="lineno"> 83</span> constant <span class="keyword">const</span> <span class="keywordtype">int</span>* shape,</div>
|
||||
<div class="line"><a id="l00084" name="l00084"></a><span class="lineno"> 84</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* a_strides,</div>
|
||||
<div class="line"><a id="l00085" name="l00085"></a><span class="lineno"> 85</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* b_strides,</div>
|
||||
<div class="line"><a id="l00086" name="l00086"></a><span class="lineno"> 86</span> constant <span class="keyword">const</span> <span class="keywordtype">size_t</span>* c_strides,</div>
|
||||
<div class="line"><a id="l00087" name="l00087"></a><span class="lineno"> 87</span> constant <span class="keyword">const</span> <span class="keywordtype">int</span>& ndim,</div>
|
||||
<div class="line"><a id="l00088" name="l00088"></a><span class="lineno"> 88</span> uint3 index [[thread_position_in_grid]],</div>
|
||||
<div class="line"><a id="l00089" name="l00089"></a><span class="lineno"> 89</span> uint3 grid_dim [[threads_per_grid]]) {</div>
|
||||
<div class="line"><a id="l00090" name="l00090"></a><span class="lineno"> 90</span> <span class="keyword">auto</span> idx = <a class="code hl_function" href="backend_2metal_2kernels_2utils_8h.html#a65d87b425e1f8ca19df97c15049f8733">elem_to_loc_3_nd<IdxT></a>(</div>
|
||||
<div class="line"><a id="l00091" name="l00091"></a><span class="lineno"> 91</span> {N * index.x, index.y, index.z},</div>
|
||||
<div class="line"><a id="l00092" name="l00092"></a><span class="lineno"> 92</span> shape,</div>
|
||||
<div class="line"><a id="l00093" name="l00093"></a><span class="lineno"> 93</span> a_strides,</div>
|
||||
<div class="line"><a id="l00094" name="l00094"></a><span class="lineno"> 94</span> b_strides,</div>
|
||||
<div class="line"><a id="l00095" name="l00095"></a><span class="lineno"> 95</span> c_strides,</div>
|
||||
<div class="line"><a id="l00096" name="l00096"></a><span class="lineno"> 96</span> ndim);</div>
|
||||
<div class="line"><a id="l00097" name="l00097"></a><span class="lineno"> 97</span> <span class="keyword">auto</span> xshape = shape[ndim - 1];</div>
|
||||
<div class="line"><a id="l00098" name="l00098"></a><span class="lineno"> 98</span> IdxT out_idx = N * index.x + xshape * (index.y + IdxT(grid_dim.y) * index.z);</div>
|
||||
<div class="line"><a id="l00099" name="l00099"></a><span class="lineno"> 99</span> IdxT a_xstride = a_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00100" name="l00100"></a><span class="lineno"> 100</span> IdxT b_xstride = b_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00101" name="l00101"></a><span class="lineno"> 101</span> IdxT c_xstride = c_strides[ndim - 1];</div>
|
||||
<div class="line"><a id="l00102" name="l00102"></a><span class="lineno"> 102</span> <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i < N && (int(N * index.x) + i) < xshape; ++i) {</div>
|
||||
<div class="line"><a id="l00103" name="l00103"></a><span class="lineno"> 103</span> d[out_idx++] = Op()(a[idx.x], b[idx.y], c[idx.z]);</div>
|
||||
<div class="line"><a id="l00104" name="l00104"></a><span class="lineno"> 104</span> idx.x += a_xstride;</div>
|
||||
<div class="line"><a id="l00105" name="l00105"></a><span class="lineno"> 105</span> idx.y += b_xstride;</div>
|
||||
<div class="line"><a id="l00106" name="l00106"></a><span class="lineno"> 106</span> idx.z += c_xstride;</div>
|
||||
<div class="line"><a id="l00107" name="l00107"></a><span class="lineno"> 107</span> }</div>
|
||||
<div class="line"><a id="l00108" name="l00108"></a><span class="lineno"> 108</span>}</div>
|
||||
</div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a196a07022b812b241d4c06192c0fa83d"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a196a07022b812b241d4c06192c0fa83d">elem_to_loc_1</a></div><div class="ttdeci">METAL_FUNC stride_t elem_to_loc_1(uint elem, constant const stride_t &stride)</div><div class="ttdef"><b>Definition</b> utils.h:133</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a2c34ed54714c69e6e1b44344f9e6e330"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a2c34ed54714c69e6e1b44344f9e6e330">elem_to_loc_3</a></div><div class="ttdeci">METAL_FUNC stride_t elem_to_loc_3(uint3 elem, constant const stride_t strides[3])</div><div class="ttdef"><b>Definition</b> utils.h:145</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a66940b1cc3d64651d24634bc696d528b"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a66940b1cc3d64651d24634bc696d528b">elem_to_loc_3_nd</a></div><div class="ttdeci">METAL_FUNC ulong3 elem_to_loc_3_nd(uint3 elem, constant const int *shape, constant const size_t *a_strides, constant const size_t *b_strides, constant const size_t *c_strides, int ndim)</div><div class="ttdef"><b>Definition</b> utils.h:171</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_ad6c45cacca97899cd362df49c06fea79"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#ad6c45cacca97899cd362df49c06fea79">elem_to_loc_2</a></div><div class="ttdeci">METAL_FUNC stride_t elem_to_loc_2(uint2 elem, constant const stride_t strides[2])</div><div class="ttdef"><b>Definition</b> utils.h:139</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_a113df0c8a841b0e986900d580644e047"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#a113df0c8a841b0e986900d580644e047">ternary_g_nd3</a></div><div class="ttdeci">void ternary_g_nd3(device const bool *a, device const T *b, device const T *c, device T *d, constant const size_t a_strides[3], constant const size_t b_strides[3], constant const size_t c_strides[3], uint3 index, uint3 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:60</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a43f33efc000962d6de881a3aab7458de"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a43f33efc000962d6de881a3aab7458de">elem_to_loc_2</a></div><div class="ttdeci">METAL_FUNC IdxT elem_to_loc_2(uint2 elem, constant const StrideT strides[2])</div><div class="ttdef"><b>Definition</b> utils.h:145</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a650f8ea8cf9f9519da9e301aad0308dc"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a650f8ea8cf9f9519da9e301aad0308dc">elem_to_loc_3</a></div><div class="ttdeci">METAL_FUNC IdxT elem_to_loc_3(uint3 elem, constant const StrideT strides[3])</div><div class="ttdef"><b>Definition</b> utils.h:150</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_a65d87b425e1f8ca19df97c15049f8733"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#a65d87b425e1f8ca19df97c15049f8733">elem_to_loc_3_nd</a></div><div class="ttdeci">METAL_FUNC vec< IdxT, 3 > elem_to_loc_3_nd(uint3 elem, constant const int *shape, constant const size_t *a_strides, constant const size_t *b_strides, constant const size_t *c_strides, int ndim)</div><div class="ttdef"><b>Definition</b> utils.h:182</div></div>
|
||||
<div class="ttc" id="abackend_2metal_2kernels_2utils_8h_html_ac612d0ae30b8257198339debe04916a3"><div class="ttname"><a href="backend_2metal_2kernels_2utils_8h.html#ac612d0ae30b8257198339debe04916a3">elem_to_loc_1</a></div><div class="ttdeci">METAL_FUNC IdxT elem_to_loc_1(uint elem, constant const StrideT &stride)</div><div class="ttdef"><b>Definition</b> utils.h:140</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_a046dcbf67cd2318d45355dc7516e3ff4"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#a046dcbf67cd2318d45355dc7516e3ff4">ternary_g_nd3</a></div><div class="ttdeci">void ternary_g_nd3(device const bool *a, device const T *b, device const T *c, device T *d, constant const size_t a_strides[3], constant const size_t b_strides[3], constant const size_t c_strides[3], uint3 index, uint3 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:60</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_a1bd5918559850f3f80e3adee2391fe6a"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#a1bd5918559850f3f80e3adee2391fe6a">ternary_g_nd1</a></div><div class="ttdeci">void ternary_g_nd1(device const bool *a, device const T *b, device const T *c, device T *d, constant const size_t &a_strides, constant const size_t &b_strides, constant const size_t &c_strides, uint index)</div><div class="ttdef"><b>Definition</b> ternary.h:26</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_a3e610f3b01966bdbf23fdfebe5d2c508"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#a3e610f3b01966bdbf23fdfebe5d2c508">ternary_v2</a></div><div class="ttdeci">void ternary_v2(device const bool *a, device const T *b, device const T *c, device T *d, uint2 index, uint2 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:14</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_a83f93644d21ee774e06e8190d0725ccb"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#a83f93644d21ee774e06e8190d0725ccb">ternary_v</a></div><div class="ttdeci">void ternary_v(device const bool *a, device const T *b, device const T *c, device T *d, uint index)</div><div class="ttdef"><b>Definition</b> ternary.h:4</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_adf8b5989de971e43829875dc0097cdfb"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#adf8b5989de971e43829875dc0097cdfb">ternary_g</a></div><div class="ttdeci">void ternary_g(device const bool *a, device const T *b, device const T *c, device T *d, constant const int *shape, constant const size_t *a_strides, constant const size_t *b_strides, constant const size_t *c_strides, constant const int &ndim, uint3 index, uint3 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:79</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_afdf0d9d0cb21fcb3f176500785076af8"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#afdf0d9d0cb21fcb3f176500785076af8">ternary_g_nd2</a></div><div class="ttdeci">void ternary_g_nd2(device const bool *a, device const T *b, device const T *c, device T *d, constant const size_t a_strides[2], constant const size_t b_strides[2], constant const size_t c_strides[2], uint2 index, uint2 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:42</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_ab2051fd944c2e24c57d5b4af54894d72"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#ab2051fd944c2e24c57d5b4af54894d72">ternary_g</a></div><div class="ttdeci">void ternary_g(device const bool *a, device const T *b, device const T *c, device T *d, constant const int *shape, constant const size_t *a_strides, constant const size_t *b_strides, constant const size_t *c_strides, constant const int &ndim, uint3 index, uint3 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:78</div></div>
|
||||
<div class="ttc" id="ametal_2kernels_2ternary_8h_html_adec9ca8a8bf527cb15d70da5857af15d"><div class="ttname"><a href="metal_2kernels_2ternary_8h.html#adec9ca8a8bf527cb15d70da5857af15d">ternary_g_nd2</a></div><div class="ttdeci">void ternary_g_nd2(device const bool *a, device const T *b, device const T *c, device T *d, constant const size_t a_strides[2], constant const size_t b_strides[2], constant const size_t c_strides[2], uint2 index, uint2 grid_dim)</div><div class="ttdef"><b>Definition</b> ternary.h:42</div></div>
|
||||
</div><!-- fragment --></div><!-- contents -->
|
||||
<!-- start footer part -->
|
||||
<hr class="footer"/><address class="footer"><small>
|
||||
|
||||
Reference in New Issue
Block a user