github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-07 03:00:26 -04:00
Code Issues Packages Projects Releases Wiki Activity

qcom fix image pitch (#6600)

Browse Source 
* qcom fix image pitch

* correct
This commit is contained in:
nimlgen
2024-09-19 18:50:02 +08:00
committed by GitHub
parent a1a882b006
commit 944cc46e11
1 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
tinygrad/runtime/ops_qcom.py
View File

@@ -279,8 +279,15 @@ class QCOMProgram(HCQProgram):
class QCOMAllocator(HCQAllocator):
def _alloc(self, size:int, options:BufferOptions) -> HCQBuffer:
if options.image is not None:
pitch = round_up(round_up(options.image.shape[1], 16) * (4 * options.image.base.itemsize), 1 << (pitchalign:=6))
texture = self.device._gpu_alloc(pitch * round_up(options.image.shape[0], 16), kgsl.KGSL_MEMTYPE_TEXTURE, map_to_cpu=True)
imgw, imgh, itemsize_log = options.image.shape[1], options.image.shape[0], int(math.log2(options.image.itemsize))
pitchalign = max(6, 11 - int(math.log2(imgh))) if imgh > 1 else 6
align_up = max(1, (8 // itemsize_log + 1) - imgh // 32) if pitchalign == 6 else (2 ** (pitchalign - itemsize_log - 2))
granularity = 128 if options.image.itemsize == 4 else 256
pitch_add = (1 << pitchalign) if min(next_power2(imgw), round_up(imgw, granularity)) - align_up + 1 <= imgw and imgw > granularity//2 else 0
pitch = round_up(imgw * 4 * options.image.itemsize, 1 << pitchalign) + pitch_add
texture = self.device._gpu_alloc(pitch * round_up(imgh, 16), kgsl.KGSL_MEMTYPE_TEXTURE, map_to_cpu=True)
# Extend HCQBuffer with texture-related info.
texture.samplers, texture.descriptor, texture.ibo = [0] * 4, [0] * 16, [0] * 16
@@ -291,7 +298,7 @@ class QCOMAllocator(HCQAllocator):
tex_fmt = adreno.FMT6_32_32_32_32_FLOAT if options.image.itemsize == 4 else adreno.FMT6_16_16_16_16_FLOAT
texture.descriptor[0] = qreg.a6xx_tex_const_0(swiz_x=0, swiz_y=1, swiz_z=2, swiz_w=3, fmt=tex_fmt)
texture.descriptor[1] = qreg.a6xx_tex_const_1(width=options.image.shape[1], height=options.image.shape[0])
texture.descriptor[1] = qreg.a6xx_tex_const_1(width=imgw, height=imgh)
texture.descriptor[2] = qreg.a6xx_tex_const_2(type=adreno.A6XX_TEX_2D, pitch=pitch, pitchalign=pitchalign-6)
texture.descriptor[4:7] = [*data64_le(texture.va_addr), qreg.a6xx_tex_const_6(plane_pitch=0x400000)]
texture.ibo = [texture.descriptor[0] & (~0xffff), *texture.descriptor[1:len(texture.descriptor)]]