From b1c5e21fb5dc9eb10f18abbf66153dc17ee9c5b5 Mon Sep 17 00:00:00 2001
From: Michael Kuron <mkuron@icp.uni-stuttgart.de>
Date: Thu, 6 Apr 2023 08:27:06 +0200
Subject: [PATCH] Remove support for non-power-of-2 SVE vector widths
---
.gitlab-ci.yml | 2 +-
pystencils/backends/arm_instruction_sets.py | 6 ++----
pystencils/backends/simd_instruction_sets.py | 14 +++++---------
3 files changed, 8 insertions(+), 14 deletions(-)
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index e50894bd6..f333e761d 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -175,7 +175,7 @@ arm64v9:
extends: .multiarch_template
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/arm64
variables:
- PYSTENCILS_SIMD: "sve256,sve512,sve"
+ PYSTENCILS_SIMD: "sve128,sve256,sve512,sve"
before_script:
- *multiarch_before_script
- sed -i s/march=native/march=armv8-a+sve/g ~/.config/pystencils/config.json
diff --git a/pystencils/backends/arm_instruction_sets.py b/pystencils/backends/arm_instruction_sets.py
index 73ea7eb44..9aa8f6c0a 100644
--- a/pystencils/backends/arm_instruction_sets.py
+++ b/pystencils/backends/arm_instruction_sets.py
@@ -151,9 +151,7 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
result['any'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) > 0'
result['all'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) == 16*0xff'
- if instruction_set == 'sve' or bitwidth & (bitwidth - 1) == 0:
- # only power-of-2 vector sizes will evenly divide a cacheline
- result['cachelineSize'] = 'cachelineSize()'
- result['cachelineZero'] = 'cachelineZero((void*) {0})'
+ result['cachelineSize'] = 'cachelineSize()'
+ result['cachelineZero'] = 'cachelineZero((void*) {0})'
return result
diff --git a/pystencils/backends/simd_instruction_sets.py b/pystencils/backends/simd_instruction_sets.py
index cdb2ee5cf..7d0d028c0 100644
--- a/pystencils/backends/simd_instruction_sets.py
+++ b/pystencils/backends/simd_instruction_sets.py
@@ -1,4 +1,3 @@
-import math
import os
import platform
from ctypes import CDLL
@@ -86,15 +85,12 @@ def get_supported_instruction_sets():
if flags.issuperset(required_sve_flags):
if platform.system() == 'Linux':
libc = CDLL('libc.so.6')
- native_length = 8 * libc.prctl(51, 0, 0, 0, 0) # PR_SVE_GET_VL
- if native_length < 0:
+ length = 8 * libc.prctl(51, 0, 0, 0, 0) # PR_SVE_GET_VL
+ if length < 0:
raise OSError("SVE length query failed")
- pwr2_length = int(2**math.floor(math.log2(native_length)))
- if pwr2_length % 256 == 0:
- result.append(f"sve{pwr2_length//2}")
- if native_length != pwr2_length:
- result.append(f"sve{pwr2_length}")
- result.append(f"sve{native_length}")
+ while length > 128:
+ result.append(f"sve{length}")
+ length //= 2
result.append("sve")
return result
--
GitLab