* Add nvptx pass * Fix spaces * Don't change name * Add runtime support * Add init call * Add more runtime functions * Add launch function * Add intrinsics * Fix codegen * Run GPU pass between general opt passes * Set data layout * Create context * Link libdevice * Add function remapping * Fix linkage * Fix libdevice link * Fix linking * Fix personality * Fix linking * Fix linking * Fix linking * Add internalize pass * Add more math conversions * Add more re-mappings * Fix conversions * Fix __str__ * Add decorator attribute for any decorator * Update kernel decorator * Fix kernel decorator * Fix kernel decorator * Fix kernel decorator * Fix kernel decorator * Remove old decorator * Fix pointer calc * Fix fill-in codegen * Fix linkage * Add comment * Update list conversion * Add more conversions * Add dict and set conversions * Add float32 type to IR/LLVM * Add float32 * Add float32 stdlib * Keep required global values in PTX module * Fix PTX module pruning * Fix malloc * Set will-return * Fix name cleanup * Fix access * Fix name cleanup * Fix function renaming * Update dimension API * Fix args * Clean up API * Move GPU transformations to end of opt pipeline * Fix alloc replacements * Fix naming * Target PTX 4.2 * Fix global renaming * Fix early return in static blocks; Add __realized__ function * Format * Add __llvm_name__ for functions * Add vector type to IR * SIMD support [wip] * Update kernel naming * Fix early returns; Fix SIMD calls * Fix kernel naming * Fix IR matcher * Remove module print * Update realloc * Add overloads for 32-bit float math ops * Add gpu.Pointer type for working with raw pointers * Add float32 conversion * Add to_gpu and from_gpu * clang-format * Add f32 reduction support to OpenMP * Fix automatic GPU class conversions * Fix conversion functions * Fix conversions * Rename self * Fix tuple conversion * Fix conversions * Fix conversions * Update PTX filename * Fix filename * Add raw function * Add GPU docs * Allow nested object conversions * Add tests (WIP) * Update SIMD * Add staticrange and statictuple loop support * SIMD updates * Add new Vec constructors * Fix UInt conversion * Fix size-0 allocs * Add more tests * Add matmul test * Rename gpu test file * Add more tests * Add alloc cache * Fix object_to_gpu * Fix frees * Fix str conversion * Fix set conversion * Fix conversions * Fix class conversion * Fix str conversion * Fix byte conversion * Fix list conversion * Fix pointer conversions * Fix conversions * Fix conversions * Update tests * Fix conversions * Fix tuple conversion * Fix tuple conversion * Fix auto conversions * Fix conversion * Fix magics * Update tests * Support GPU in JIT mode * Fix GPU+JIT * Fix kernel filename in JIT mode * Add __static_print__; Add earlyDefines; Various domination bugfixes; SimplifyContext RAII base handling * Fix global static handling * Fix float32 tests * FIx gpu module * Support OpenMP "collapse" option * Add more collapse tests * Capture generics and statics * TraitVar handling * Python exceptions / isinstance [wip; no_ci] * clang-format * Add list comparison operators * Support empty raise in IR * Add dict 'or' operator * Fix repr * Add copy module * Fix spacing * Use sm_30 * Python exceptions * TypeTrait support; Fix defaultDict * Fix earlyDefines * Add defaultdict * clang-format * Fix invalid canonicalizations * Fix empty raise * Fix copyright * Add Python numerics option * Support py-numerics in math module * Update docs * Add static Python division / modulus * Add static py numerics tests * Fix staticrange/tuple; Add KwTuple.__getitem__ * clang-format * Add gpu parameter to par * Fix globals * Don't init loop vars on loop collapse * Add par-gpu tests * Update gpu docs * Fix isinstance check * Remove invalid test * Add -libdevice to set custom path [skip ci] * Add release notes; bump version [skip ci] * Add libdevice docs [skip ci] Co-authored-by: Ibrahim Numanagić <ibrahimpasa@gmail.com>
7.6 KiB
Codon supports GPU programming through a native GPU backend. Currently, only Nvidia devices are supported. Here is a simple example:
import gpu
@gpu.kernel
def hello(a, b, c):
i = gpu.thread.x
c[i] = a[i] + b[i]
a = [i for i in range(16)]
b = [2*i for i in range(16)]
c = [0 for _ in range(16)]
hello(a, b, c, grid=1, block=16)
print(c)
which outputs:
[0, 3, 6, 9, 12, 15, 18, 21, 24, 27, 30, 33, 36, 39, 42, 45]
The same code can be written using Codon's @par
syntax:
a = [i for i in range(16)]
b = [2*i for i in range(16)]
c = [0 for _ in range(16)]
@par(gpu=True)
for i in range(16):
c[i] = a[i] + b[i]
print(c)
Below is a more comprehensive example for computing the Mandelbrot set, and plotting it using NumPy/Matplotlib:
from python import numpy as np
from python import matplotlib.pyplot as plt
import gpu
MAX = 1000 # maximum Mandelbrot iterations
N = 4096 # width and height of image
pixels = [0 for _ in range(N * N)]
def scale(x, a, b):
return a + (x/N)*(b - a)
@gpu.kernel
def mandelbrot(pixels):
idx = (gpu.block.x * gpu.block.dim.x) + gpu.thread.x
i, j = divmod(idx, N)
c = complex(scale(j, -2.00, 0.47), scale(i, -1.12, 1.12))
z = 0j
iteration = 0
while abs(z) <= 2 and iteration < MAX:
z = z**2 + c
iteration += 1
pixels[idx] = int(255 * iteration/MAX)
mandelbrot(pixels, grid=(N*N)//1024, block=1024)
plt.imshow(np.array(pixels).reshape(N, N))
plt.show()
The GPU version of the Mandelbrot code is about 450 times faster than an equivalent CPU version.
GPU kernels are marked with the @gpu.kernel
annotation, and
compiled specially in Codon's backend. Kernel functions can
use the vast majority of features supported in Codon, with a
couple notable exceptions:
-
Exception handling is not supported inside the kernel, meaning kernel code should not throw or catch exceptions.
raise
statements inside the kernel are marked as unreachable and optimized out. -
Functionality related to I/O is not supported (e.g. you can't open a file in the kernel).
-
A few other modules and functions are not allowed, such as the
re
module (which uses an external regex library) or theos
module.
{% hint style="warning" %} The GPU module is under active development. APIs and semantics might change between Codon releases. {% endhint %}
Invoking the kernel
The kernel can be invoked via a simple call with added grid
and
block
parameters. These parameters define the grid and block
dimensions, respectively. Recall that GPU execution involves a grid
of (X
x Y
x Z
) blocks where each block contains (x
x y
x z
)
executing threads. Device-specific restrictions on grid and block sizes
apply.
The grid
and block
parameters can be one of:
- Single integer
x
, giving dimensions(x, 1, 1)
- Tuple of two integers
(x, y)
, giving dimensions(x, y, 1)
- Tuple of three integers
(x, y, z)
, giving dimensions(x, y, z)
- Instance of
gpu.Dim3
as inDim3(x, y, z)
, specifying the three dimensions
GPU intrinsics
Codon's GPU module provides many of the same intrinsics that CUDA does:
Codon | Description | CUDA equivalent |
---|---|---|
gpu.thread.x |
x-coordinate of current thread in block | threadId.x |
gpu.block.x |
x-coordinate of current block in grid | blockIdx.x |
gpu.block.dim.x |
x-dimension of block | blockDim.x |
gpu.grid.dim.x |
x-dimension of grid | gridDim.x |
The same applies for the y
and z
coordinates. The *.dim
objects are instances
of gpu.Dim3
.
Math functions
All the functions in the math
module are supported in kernel functions, and
are automatically replaced with GPU-optimized versions:
import math
import gpu
@gpu.kernel
def hello(x):
i = gpu.thread.x
x[i] = math.sqrt(x[i]) # uses __nv_sqrt from libdevice
x = [float(i) for i in range(10)]
hello(x, grid=1, block=10)
print(x)
gives:
[0, 1, 1.41421, 1.73205, 2, 2.23607, 2.44949, 2.64575, 2.82843, 3]
Libdevice
Codon uses libdevice
for GPU-optimized math functions. The default libdevice path is
/usr/local/cuda/nvvm/libdevice/libdevice.10.bc
. An alternative path can be specified
via the -libdevice
compiler flag.
Working with raw pointers
By default, objects are converted entirely to their GPU counterparts, which have the same data layout as the original objects (although the Codon compiler might perform optimizations by swapping a CPU implementation of a data type with a GPU-optimized implementation that exposes the same API). This preserves all of Codon/Python's standard semantics within the kernel.
It is possible to use a kernel with raw pointers via gpu.raw
, which corresponds
to how the kernel would be written in C++/CUDA:
import gpu
@gpu.kernel
def hello(a, b, c):
i = gpu.thread.x
c[i] = a[i] + b[i]
a = [i for i in range(16)]
b = [2*i for i in range(16)]
c = [0 for _ in range(16)]
# call the kernel with three int-pointer arguments:
hello(gpu.raw(a), gpu.raw(b), gpu.raw(c), grid=1, block=16)
print(c) # output same as first snippet's
gpu.raw
can avoid an extra pointer indirection, but outputs a Codon Ptr
object,
meaning the corresponding kernel parameters will not have the full list API, instead
having the more limited Ptr
API (which primarily just supports indexing/assignment).
Object conversions
A hidden API is used to copy objects to and from the GPU device. This API consists of two new magic methods:
-
__to_gpu__(self)
: Allocates the necessary GPU memory and copies the objectself
to the device. -
__from_gpu__(self, gpu_object)
: Copies the GPU memory ofgpu_object
(which is a value returned by__to_gpu__
) back to the CPU objectself
.
For primitive types like int
and float
, __to_gpu__
simply returns self
and
__from_gpu__
does nothing. These methods are defined for all the built-in types and
are automatically generated for user-defined classes, so most objects can be transferred
back and forth from the GPU seamlessly. A user-defined class that makes use of raw pointers
or other low-level constructs will have to define these methods for GPU use. Please refer
to the gpu
module for implementation examples.
@par(gpu=True)
Codon's @par
syntax can be used to seamlessly parallelize existing loops on the GPU,
without needing to explicitly write them as kernels. For loop nests, the collapse
argument
can be used to cover the entire iteration space on the GPU. For example, here is the Mandelbrot
code above written using @par
:
MAX = 1000 # maximum Mandelbrot iterations
N = 4096 # width and height of image
pixels = [0 for _ in range(N * N)]
def scale(x, a, b):
return a + (x/N)*(b - a)
@par(gpu=True, collapse=2)
for i in range(N):
for j in range(N):
c = complex(scale(j, -2.00, 0.47), scale(i, -1.12, 1.12))
z = 0j
iteration = 0
while abs(z) <= 2 and iteration < MAX:
z = z**2 + c
iteration += 1
pixels[i*N + j] = int(255 * iteration/MAX)
Note that the gpu=True
option disallows shared variables (i.e. assigning out-of-loop
variables in the loop body) as well as reductions. The other GPU-specific restrictions
described here apply as well.
Troubleshooting
CUDA errors resulting in kernel abortion are printed, and typically arise from invalid code in the kernel, either via using exceptions or using unsupported modules/objects.