mirror of
https://github.com/vladmandic/sdnext.git
synced 2026-10-12 02:24:13 +00:00
Logging options & UI Cleanup, useless option removal, minimizing on options never used
This commit is contained in:
resonantsky
@@ -9,7 +9,7 @@ from modules.logger import log
|
||||
from modules.json_helpers import readfile, writefile
|
||||
from modules.shared import opts
|
||||
|
||||
from scripts.rocm.rocm_vars import ROCM_ENV_VARS # pylint: disable=no-name-in-module
|
||||
from scripts.rocm.rocm_vars import ROCM_ENV_VARS, MIOPEN_LOGGING_VARS # pylint: disable=no-name-in-module
|
||||
from scripts.rocm import rocm_profiles # pylint: disable=no-name-in-module
|
||||
if sys.platform == "win32":
|
||||
from scripts.rocm import rocm_log_win32 as rocm_log # pylint: disable=no-name-in-module
|
||||
@@ -207,6 +207,25 @@ def _dropdown_choices(options):
|
||||
return options
|
||||
|
||||
|
||||
def _set_miopen_logging(config: Dict[str, str], enabled: bool) -> None:
|
||||
config["MIOPEN_LOG_LEVEL"] = "5" if enabled else "0"
|
||||
config["MIOPEN_DEBUG_ENABLE"] = "1" if enabled else "0"
|
||||
|
||||
|
||||
def _normalize_miopen_logging(config: Dict[str, str]) -> bool:
|
||||
"""Keep both MIOpen logging variables in sync; DEBUG_ENABLE is authoritative."""
|
||||
enabled = str(config.get("MIOPEN_DEBUG_ENABLE", ROCM_ENV_VARS["MIOPEN_DEBUG_ENABLE"]["default"])) == "1"
|
||||
previous = (config.get("MIOPEN_LOG_LEVEL"), config.get("MIOPEN_DEBUG_ENABLE"))
|
||||
_set_miopen_logging(config, enabled)
|
||||
return previous != (config["MIOPEN_LOG_LEVEL"], config["MIOPEN_DEBUG_ENABLE"])
|
||||
|
||||
|
||||
def _apply_miopen_logging_env(config: Dict[str, str]) -> None:
|
||||
_normalize_miopen_logging(config)
|
||||
for name in MIOPEN_LOGGING_VARS:
|
||||
os.environ[name] = config[name]
|
||||
|
||||
|
||||
# --- config I/O ---
|
||||
|
||||
def load_config() -> Dict[str, str]:
|
||||
@@ -222,10 +241,13 @@ def load_config() -> Dict[str, str]:
|
||||
dirty = {k for k in _cache if k in _UNSET_VARS or (k != _ARCH_KEY and k not in ROCM_ENV_VARS)}
|
||||
if dirty:
|
||||
_cache = {k: v for k, v in _cache.items() if k not in dirty}
|
||||
logging_changed = _normalize_miopen_logging(_cache)
|
||||
if dirty or logging_changed:
|
||||
writefile(_cache, str(CONFIG))
|
||||
log.debug(f'ROCm: config={str(CONFIG)} purged={len(dirty)} stale/unsafe')
|
||||
log.debug(f'ROCm: config={str(CONFIG)} purged={len(dirty)} stale/unsafe miopen_logging_normalized={logging_changed}')
|
||||
else:
|
||||
_cache = {k: v["default"] for k, v in ROCM_ENV_VARS.items()}
|
||||
_normalize_miopen_logging(_cache)
|
||||
log.debug(f'ROCm: load config={str(CONFIG)} items={len(_cache)}')
|
||||
return _cache
|
||||
|
||||
@@ -233,6 +255,7 @@ def load_config() -> Dict[str, str]:
|
||||
def save_config(config: Dict[str, str]) -> None:
|
||||
global _cache # pylint: disable=global-statement
|
||||
sanitized = {k: v for k, v in config.items() if k not in _UNSET_VARS}
|
||||
_normalize_miopen_logging(sanitized)
|
||||
# Enforce arch-incompatible solvers to "0" before writing.
|
||||
# Prevents malformed edits (UI or JSON hand-edit) from persisting incompatible "1" values.
|
||||
arch = sanitized.get(_ARCH_KEY, "")
|
||||
@@ -246,8 +269,8 @@ def save_config(config: Dict[str, str]) -> None:
|
||||
|
||||
|
||||
def apply_env(config: Optional[Dict[str, str]] = None) -> None:
|
||||
if config is None:
|
||||
config = load_config()
|
||||
config = dict(config if config is not None else load_config())
|
||||
_normalize_miopen_logging(config)
|
||||
for var in _UNSET_VARS | _EXTRA_CLEAR_VARS:
|
||||
if var in os.environ:
|
||||
del os.environ[var]
|
||||
@@ -285,7 +308,7 @@ def stop_miopen_logging() -> None:
|
||||
rocm_log.stop_miopen_logging()
|
||||
|
||||
|
||||
def apply_all(names: list, values: list) -> None:
|
||||
def apply_all(names: list, values: list, miopen_logging: Optional[bool] = None) -> None:
|
||||
config = load_config().copy()
|
||||
arch = config.get(_ARCH_KEY, "")
|
||||
unavailable = rocm_profiles.UNAVAILABLE.get(arch, set())
|
||||
@@ -314,6 +337,16 @@ def apply_all(names: list, values: list) -> None:
|
||||
if meta.get("options"):
|
||||
value = _dropdown_stored(str(value), meta["options"])
|
||||
config[name] = _collapse_venv(str(value))
|
||||
if miopen_logging is not None:
|
||||
_set_miopen_logging(config, miopen_logging)
|
||||
save_config(config)
|
||||
apply_env(config)
|
||||
|
||||
|
||||
def set_miopen_logging(enabled: bool) -> None:
|
||||
"""Persist and apply both MIOpen logging environment variables together."""
|
||||
config = load_config().copy()
|
||||
_set_miopen_logging(config, enabled)
|
||||
save_config(config)
|
||||
apply_env(config)
|
||||
|
||||
@@ -465,7 +498,7 @@ def info() -> dict:
|
||||
|
||||
# --- ROCm / HIP package versions ---
|
||||
rocm_pkgs = {}
|
||||
for pkg in ("rocm", "rocm-sdk-core", "rocm-sdk-devel"):
|
||||
for pkg in ("rocm",):
|
||||
v = _pkg_version(pkg)
|
||||
if v != "n/a":
|
||||
rocm_pkgs[pkg] = v
|
||||
@@ -556,6 +589,8 @@ if _is_rocm_runtime():
|
||||
try:
|
||||
if CONFIG.exists():
|
||||
apply_env()
|
||||
else:
|
||||
_apply_miopen_logging_env(load_config())
|
||||
rocm_log.start_miopen_logging()
|
||||
except Exception as _e:
|
||||
log.debug(f"[rocm_mgr] Warning: failed to apply env at import: {_e}")
|
||||
|
||||
+164
-157
@@ -18,6 +18,20 @@ def _sitepackages_subpath(*parts: str) -> str: # auto-path helper (currently not
|
||||
|
||||
|
||||
GENERAL_VARS: Dict[str, Dict[str, Any]] = {
|
||||
"MIOPEN_FIND_MODE": {
|
||||
"default": "2",
|
||||
"desc": "MIOpen Find Mode",
|
||||
"widget": "dropdown",
|
||||
"options": [("1 - NORMAL", "1"), ("2 - FAST", "2"), ("3 - HYBRID", "3"), ("5 - DYNAMIC_HYBRID", "5"), ("6 - TRUST_VERIFY", "6"), ("7 - TRUST_VERIFY_FULL", "7")],
|
||||
"restart_required": True,
|
||||
},
|
||||
"MIOPEN_FIND_ENFORCE": {
|
||||
"default": "1",
|
||||
"desc": "MIOpen Find Enforce",
|
||||
"widget": "dropdown",
|
||||
"options": [("1 - NONE", "1"), ("2 - DB_UPDATE", "2"), ("3 - SEARCH", "3"), ("4 - SEARCH_DB_UPDATE", "4"), ("5 - DB_CLEAN", "5")],
|
||||
"restart_required": True,
|
||||
},
|
||||
"MIOPEN_SYSTEM_DB_PATH": {
|
||||
# "default": _sitepackages_subpath("{LIBS_PKG}", _BIN_DIR) + os.sep, # auto-path disabled; kept for troubleshooting.
|
||||
"default": "",
|
||||
@@ -34,6 +48,20 @@ GENERAL_VARS: Dict[str, Dict[str, Any]] = {
|
||||
"options": None,
|
||||
"restart_required": True,
|
||||
},
|
||||
"MIOPEN_LOG_LEVEL": {
|
||||
"default": "0",
|
||||
"desc": "MIOpen log verbosity level",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Default", "0"), ("1 - Quiet", "1"), ("3 - Error", "3"), ("4 - Warning", "4"), ("5 - Info", "5"), ("6 - Detail", "6"), ("7 - Trace", "7")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"MIOPEN_DEBUG_ENABLE": {
|
||||
"default": "0",
|
||||
"desc": "Enable MIOpen logging",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"MIOPEN_GEMM_ENFORCE_BACKEND": {
|
||||
"default": "1",
|
||||
"desc": "GEMM backend",
|
||||
@@ -62,20 +90,7 @@ GENERAL_VARS: Dict[str, Dict[str, Any]] = {
|
||||
"options": [("0 - Tensile (rocBLAS)", "0"), ("1 - hipBLASLt", "1")],
|
||||
"restart_required": True,
|
||||
},
|
||||
"MIOPEN_FIND_MODE": {
|
||||
"default": "2",
|
||||
"desc": "MIOpen Find Mode",
|
||||
"widget": "dropdown",
|
||||
"options": [("1 - NORMAL", "1"), ("2 - FAST", "2"), ("3 - HYBRID", "3"), ("5 - DYNAMIC_HYBRID", "5"), ("6 - TRUST_VERIFY", "6"), ("7 - TRUST_VERIFY_FULL", "7")],
|
||||
"restart_required": True,
|
||||
},
|
||||
"MIOPEN_FIND_ENFORCE": {
|
||||
"default": "1",
|
||||
"desc": "MIOpen Find Enforce",
|
||||
"widget": "dropdown",
|
||||
"options": [("1 - NONE", "1"), ("2 - DB_UPDATE", "2"), ("3 - SEARCH", "3"), ("4 - SEARCH_DB_UPDATE", "4"), ("5 - DB_CLEAN", "5")],
|
||||
"restart_required": True,
|
||||
},
|
||||
|
||||
"MIOPEN_SEARCH_CUTOFF": {
|
||||
"default": "0",
|
||||
"desc": "Enable early termination of suboptimal searches",
|
||||
@@ -97,152 +112,142 @@ GENERAL_VARS: Dict[str, Dict[str, Any]] = {
|
||||
"options": None,
|
||||
"restart_required": False,
|
||||
},
|
||||
"ROCBLAS_DEVICE_MEMORY_SIZE": {
|
||||
"default": "",
|
||||
"desc": "rocBLAS workspace size in bytes (empty = dynamic)",
|
||||
"widget": "textbox",
|
||||
"options": None,
|
||||
"restart_required": False,
|
||||
},
|
||||
"PYTORCH_TUNABLEOP_CACHE_DIR": {
|
||||
"default": os.path.join("{ROOT}", "models", "tunable"),
|
||||
"desc": "TunableOp cache directory",
|
||||
"widget": "textbox",
|
||||
"options": None,
|
||||
"restart_required": False,
|
||||
},
|
||||
|
||||
# --> OUTCOMMENTED SECTION REMOVED FROM ACTIVE ROCm CONFIGURATION REGISTRY; KEPT FOR REFERENCE <--
|
||||
|
||||
"ROCBLAS_STREAM_ORDER_ALLOC": {
|
||||
"default": "1",
|
||||
"desc": "rocBLAS stream-ordered memory allocation",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Standard", "0"), ("1 - Stream-ordered", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"ROCBLAS_DEFAULT_ATOMICS_MODE": {
|
||||
"default": "1",
|
||||
"desc": "rocBLAS allow atomics",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off (deterministic)", "0"), ("1 - On (performance)", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"PYTORCH_TUNABLEOP_ROCBLAS_ENABLED": {
|
||||
"default": "0",
|
||||
"desc": "TunableOp: Enable tuning",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"PYTORCH_TUNABLEOP_TUNING": {
|
||||
"default": "0",
|
||||
"desc": "TunableOp: Tuning mode",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Use Cache", "0"), ("1 - Benchmark new shapes", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"PYTORCH_TUNABLEOP_HIPBLASLT_ENABLED": {
|
||||
"default": "0",
|
||||
"desc": "TunableOp: benchmark hipBLASLt kernels",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"MIOPEN_LOG_LEVEL": {
|
||||
"default": "0",
|
||||
"desc": "MIOpen log verbosity level",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Default", "0"), ("1 - Quiet", "1"), ("3 - Error", "3"), ("4 - Warning", "4"), ("5 - Info", "5"), ("6 - Detail", "6"), ("7 - Trace", "7")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"MIOPEN_DEBUG_ENABLE": {
|
||||
"default": "0",
|
||||
"desc": "Enable MIOpen logging",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"ROCBLAS_LAYER": {
|
||||
"default": "0",
|
||||
"desc": "rocBLAS logging",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - Trace", "1"), ("2 - Bench", "2"), ("3 - Trace+Bench", "3"), ("4 - Profile", "4"), ("5 - Trace+Profile", "5"), ("6 - Bench+Profile", "6"), ("7 - All", "7")],
|
||||
"restart_required": False,
|
||||
},
|
||||
"HIPBLASLT_LOG_LEVEL": {
|
||||
"default": "0",
|
||||
"desc": "hipBLASLt logging",
|
||||
"widget": "dropdown",
|
||||
"options": [("0 - Off", "0"), ("1 - Error", "1"), ("2 - Trace", "2"), ("3 - Hints", "3"), ("4 - Info", "4"), ("5 - API Trace", "5")],
|
||||
"restart_required": False,
|
||||
},
|
||||
#"ROCBLAS_DEVICE_MEMORY_SIZE": {
|
||||
# "default": "",
|
||||
# "desc": "rocBLAS workspace size in bytes (empty = dynamic)",
|
||||
# "widget": "textbox",
|
||||
# "options": None,
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"PYTORCH_TUNABLEOP_CACHE_DIR": {
|
||||
# "default": os.path.join("{ROOT}", "models", "tunable"),
|
||||
# "desc": "TunableOp cache directory",
|
||||
# "widget": "textbox",
|
||||
# "options": None,
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#
|
||||
#"ROCBLAS_STREAM_ORDER_ALLOC": {
|
||||
# "default": "1",
|
||||
# "desc": "rocBLAS stream-ordered memory allocation",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Standard", "0"), ("1 - Stream-ordered", "1")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"ROCBLAS_DEFAULT_ATOMICS_MODE": {
|
||||
# "default": "1",
|
||||
# "desc": "rocBLAS allow atomics",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Off (deterministic)", "0"), ("1 - On (performance)", "1")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"PYTORCH_TUNABLEOP_ROCBLAS_ENABLED": {
|
||||
# "default": "0",
|
||||
# "desc": "TunableOp: Enable tuning",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"PYTORCH_TUNABLEOP_TUNING": {
|
||||
# "default": "0",
|
||||
# "desc": "TunableOp: Tuning mode",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Use Cache", "0"), ("1 - Benchmark new shapes", "1")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"PYTORCH_TUNABLEOP_HIPBLASLT_ENABLED": {
|
||||
# "default": "0",
|
||||
# "desc": "TunableOp: benchmark hipBLASLt kernels",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Off", "0"), ("1 - On", "1")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#
|
||||
#"ROCBLAS_LAYER": {
|
||||
# "default": "0",
|
||||
# "desc": "rocBLAS logging",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Off", "0"), ("1 - Trace", "1"), ("2 - Bench", "2"), ("3 - Trace+Bench", "3"), ("4 - Profile", "4"), ("5 - Trace+Profile", "5"), ("6 - Bench+Profile", "6"), ("7 - All", "7")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
#"HIPBLASLT_LOG_LEVEL": {
|
||||
# "default": "0",
|
||||
# "desc": "hipBLASLt logging",
|
||||
# "widget": "dropdown",
|
||||
# "options": [("0 - Off", "0"), ("1 - Error", "1"), ("2 - Trace", "2"), ("3 - Hints", "3"), ("4 - Info", "4"), ("5 - API Trace", "5")],
|
||||
# "restart_required": False,
|
||||
#},
|
||||
}
|
||||
|
||||
_SOLVER_DESCS: Dict[str, str] = {}
|
||||
|
||||
_SOLVER_DESCS.update({
|
||||
"MIOPEN_DEBUG_CONV_FFT": "Enable FFT solver",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT": "Enable Direct solver",
|
||||
"MIOPEN_DEBUG_CONV_GEMM": "Enable GEMM solver",
|
||||
"MIOPEN_DEBUG_CONV_WINOGRAD": "Enable Winograd solver",
|
||||
"MIOPEN_DEBUG_CONV_IMPLICIT_GEMM": "Enable Implicit GEMM solver",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
"MIOPEN_DEBUG_CONV_IMMED_FALLBACK": "Enable Immediate Fallback",
|
||||
"MIOPEN_DEBUG_ENABLE_AI_IMMED_MODE_FALLBACK": "Enable AI Immediate Mode Fallback",
|
||||
"MIOPEN_DEBUG_FORCE_IMMED_MODE_FALLBACK": "Force Immediate Mode Fallback",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
"MIOPEN_DEBUG_GCN_ASM_KERNELS": "Enable GCN ASM kernels",
|
||||
"MIOPEN_DEBUG_HIP_KERNELS": "Enable HIP kernels",
|
||||
"MIOPEN_DEBUG_OPENCL_CONVOLUTIONS": "Enable OpenCL convolutions",
|
||||
"MIOPEN_DEBUG_OPENCL_WAVE64_NOWGP": "Enable OpenCL Wave64 NOWGP",
|
||||
"MIOPEN_DEBUG_ATTN_SOFTMAX": "Enable Attention Softmax",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Direct ASM — FWD inference only (WRW, fixed-geometry, FP32-reference removed)
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_ASM_3X3U": "Enable Direct ASM 3x3U",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U": "Enable Direct ASM 1x1U",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1UV2": "Enable Direct ASM 1x1UV2",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U_SEARCH_OPTIMIZED": "Enable Direct ASM 1x1U Search Optimized",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U_AI_HEUR": "Enable Direct ASM 1x1U AI Heuristic",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Direct OCL — FWD inference only (WRW, FWD11X11 fixed-geom, FWDGEN FP32-ref removed)
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_OCL_FWD": "Enable Direct OCL FWD",
|
||||
"MIOPEN_DEBUG_CONV_DIRECT_OCL_FWD1X1": "Enable Direct OCL FWD1X1",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Winograd FWD — WRW removed; Fury/Rage kept as RDNA3/4 inference (off by default)
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_3X3": "Enable AMD Winograd 3x3",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RXS": "Enable AMD Winograd RxS",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RXS_FWD_BWD": "Enable AMD Winograd RxS FWD",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F3X2": "Enable AMD Winograd RxS F3x2",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F2X3": "Enable AMD Winograd RxS F2x3",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F2X3_G1": "Enable AMD Winograd RxS F2x3 G1",
|
||||
"MIOPEN_DEBUG_AMD_FUSED_WINOGRAD": "Enable AMD Fused Winograd",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_FURY_RXS_F2X3": "Enable AMD Winograd Fury RxS F2x3",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_FURY_RXS_F3X2": "Enable AMD Winograd Fury RxS F3x2",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_RAGE_RXS_F2X3": "Enable AMD Winograd Rage RxS F2x3",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Multi-pass Winograd — only F3x2/F3x3 match typical 3x3 SD shapes; wider kernels removed
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_MPASS_F3X2": "Enable AMD Winograd MPASS F3x2",
|
||||
"MIOPEN_DEBUG_AMD_WINOGRAD_MPASS_F3X3": "Enable AMD Winograd MPASS F3x3",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Implicit GEMM FWD — BWD/WRW (training), CDNA-exclusive XDLOPS variants removed
|
||||
"MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_ASM_FWD_V4R1": "Enable ASM Implicit GEMM FWD V4R1",
|
||||
"MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_ASM_FWD_V4R1_1X1": "Enable ASM Implicit GEMM FWD V4R1 1x1",
|
||||
"MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_HIP_FWD_V4R1": "Enable HIP Implicit GEMM FWD V4R1",
|
||||
"MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_HIP_FWD_V4R4": "Enable HIP Implicit GEMM FWD V4R4",
|
||||
})
|
||||
_SOLVER_DESCS.update({
|
||||
# Group Conv XDLOPS FWD — RDNA3/4 (gfx1100+) only; disabled by default
|
||||
"MIOPEN_DEBUG_GROUP_CONV_IMPLICIT_GEMM_HIP_FWD_XDLOPS": "Enable Group Conv Implicit GEMM XDLOPS FWD",
|
||||
"MIOPEN_DEBUG_GROUP_CONV_IMPLICIT_GEMM_HIP_FWD_XDLOPS_AI_HEUR": "Enable Group Conv Implicit GEMM XDLOPS FWD AI Heuristic",
|
||||
# CK (Composable Kernel) default kernels — RDNA3/4 (gfx1100+); disabled by default
|
||||
"MIOPEN_DEBUG_CK_DEFAULT_KERNELS": "Enable CK (Composable Kernel) default kernels",
|
||||
})
|
||||
# _SOLVER_DESCS: Dict[str, str] = {}
|
||||
#
|
||||
# _SOLVER_DESCS.update({
|
||||
# "MIOPEN_DEBUG_CONV_FFT": "Enable FFT solver",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT": "Enable Direct solver",
|
||||
# "MIOPEN_DEBUG_CONV_GEMM": "Enable GEMM solver",
|
||||
# "MIOPEN_DEBUG_CONV_WINOGRAD": "Enable Winograd solver",
|
||||
# "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM": "Enable Implicit GEMM solver",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# "MIOPEN_DEBUG_CONV_IMMED_FALLBACK": "Enable Immediate Fallback",
|
||||
# "MIOPEN_DEBUG_ENABLE_AI_IMMED_MODE_FALLBACK": "Enable AI Immediate Mode Fallback",
|
||||
# "MIOPEN_DEBUG_FORCE_IMMED_MODE_FALLBACK": "Force Immediate Mode Fallback",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# "MIOPEN_DEBUG_GCN_ASM_KERNELS": "Enable GCN ASM kernels",
|
||||
# "MIOPEN_DEBUG_HIP_KERNELS": "Enable HIP kernels",
|
||||
# "MIOPEN_DEBUG_OPENCL_CONVOLUTIONS": "Enable OpenCL convolutions",
|
||||
# "MIOPEN_DEBUG_OPENCL_WAVE64_NOWGP": "Enable OpenCL Wave64 NOWGP",
|
||||
# "MIOPEN_DEBUG_ATTN_SOFTMAX": "Enable Attention Softmax",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Direct ASM — FWD inference only (WRW, fixed-geometry, FP32-reference removed)
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_ASM_3X3U": "Enable Direct ASM 3x3U",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U": "Enable Direct ASM 1x1U",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1UV2": "Enable Direct ASM 1x1UV2",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U_SEARCH_OPTIMIZED": "Enable Direct ASM 1x1U Search Optimized",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_ASM_1X1U_AI_HEUR": "Enable Direct ASM 1x1U AI Heuristic",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Direct OCL — FWD inference only (WRW, FWD11X11 fixed-geom, FWDGEN FP32-ref removed)
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_OCL_FWD": "Enable Direct OCL FWD",
|
||||
# "MIOPEN_DEBUG_CONV_DIRECT_OCL_FWD1X1": "Enable Direct OCL FWD1X1",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Winograd FWD — WRW removed; Fury/Rage kept as RDNA3/4 inference (off by default)
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_3X3": "Enable AMD Winograd 3x3",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RXS": "Enable AMD Winograd RxS",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RXS_FWD_BWD": "Enable AMD Winograd RxS FWD",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F3X2": "Enable AMD Winograd RxS F3x2",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F2X3": "Enable AMD Winograd RxS F2x3",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RXS_F2X3_G1": "Enable AMD Winograd RxS F2x3 G1",
|
||||
# "MIOPEN_DEBUG_AMD_FUSED_WINOGRAD": "Enable AMD Fused Winograd",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_FURY_RXS_F2X3": "Enable AMD Winograd Fury RxS F2x3",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_FURY_RXS_F3X2": "Enable AMD Winograd Fury RxS F3x2",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_RAGE_RXS_F2X3": "Enable AMD Winograd Rage RxS F2x3",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Multi-pass Winograd — only F3x2/F3x3 match typical 3x3 SD shapes; wider kernels removed
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_MPASS_F3X2": "Enable AMD Winograd MPASS F3x2",
|
||||
# "MIOPEN_DEBUG_AMD_WINOGRAD_MPASS_F3X3": "Enable AMD Winograd MPASS F3x3",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Implicit GEMM FWD — BWD/WRW (training), CDNA-exclusive XDLOPS variants removed
|
||||
# "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_ASM_FWD_V4R1": "Enable ASM Implicit GEMM FWD V4R1",
|
||||
# "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_ASM_FWD_V4R1_1X1": "Enable ASM Implicit GEMM FWD V4R1 1x1",
|
||||
# "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_HIP_FWD_V4R1": "Enable HIP Implicit GEMM FWD V4R1",
|
||||
# "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM_HIP_FWD_V4R4": "Enable HIP Implicit GEMM FWD V4R4",
|
||||
# })
|
||||
# _SOLVER_DESCS.update({
|
||||
# # Group Conv XDLOPS FWD — RDNA3/4 (gfx1100+) only; disabled by default
|
||||
# "MIOPEN_DEBUG_GROUP_CONV_IMPLICIT_GEMM_HIP_FWD_XDLOPS": "Enable Group Conv Implicit GEMM XDLOPS FWD",
|
||||
# "MIOPEN_DEBUG_GROUP_CONV_IMPLICIT_GEMM_HIP_FWD_XDLOPS_AI_HEUR": "Enable Group Conv Implicit GEMM XDLOPS FWD AI Heuristic",
|
||||
# # CK (Composable Kernel) default kernels — RDNA3/4 (gfx1100+); disabled by default
|
||||
# "MIOPEN_DEBUG_CK_DEFAULT_KERNELS": "Enable CK (Composable Kernel) default kernels",
|
||||
# })
|
||||
|
||||
|
||||
# Solvers still in the registry but disabled by default.
|
||||
@@ -293,6 +298,8 @@ SOLVER_DTYPE_TAGS: Dict[str, str] = {
|
||||
ROCM_ENV_VARS: Dict[str, Dict[str, Any]] = {}
|
||||
ROCM_ENV_VARS.update(GENERAL_VARS)
|
||||
|
||||
MIOPEN_LOGGING_VARS = {"MIOPEN_LOG_LEVEL", "MIOPEN_DEBUG_ENABLE"}
|
||||
|
||||
SOLVER_GROUPS: List[Tuple[str, List[str]]] = []
|
||||
|
||||
# Variables that are relevant only when hipBLASLt is the active GEMM backend.
|
||||
|
||||
+28
-8
@@ -64,8 +64,6 @@ class ROCmScript(scripts_manager.Script):
|
||||
return f"<style>{' '.join(rules)}</style>" if rules else ""
|
||||
|
||||
with gr.Accordion('ROCm: Advanced Config', open=False, elem_id='rocm_config'):
|
||||
with gr.Row():
|
||||
gr.HTML("<p><u>Advanced configuration for ROCm users.</u></p><br><p>This script aims to take the guesswork out of configuring MIOpen and rocBLAS on Windows ROCm, but also to expose the functioning switches of MIOpen for advanced configurations.</p><br><p>For best performance ensure that cuDNN and PyTorch tunable ops are set to <b><i>default</i></b> in Backend Settings.</p><br><p>This script was written with the intent to support ROCm Windows users, it should however, function identically for Linux users.</p><br>")
|
||||
with gr.Row():
|
||||
btn_info = gr.Button("Refresh Info", variant="primary", elem_id="rocm_btn_info", size="sm")
|
||||
btn_apply = gr.Button("Apply", variant="primary", elem_id="rocm_btn_apply", size="sm")
|
||||
@@ -81,7 +79,16 @@ class ROCmScript(scripts_manager.Script):
|
||||
# General vars (dropdowns, textboxes, checkboxes)
|
||||
with gr.Group():
|
||||
gr.HTML("<br><h3>MIOpen Settings</h3><hr>")
|
||||
miopen_logging = "Enabled" if config.get("MIOPEN_DEBUG_ENABLE", "0") == "1" else "Disabled"
|
||||
miopen_logging_comp = gr.Dropdown(
|
||||
label="MIOpen logging",
|
||||
choices=["Disabled", "Enabled"],
|
||||
value=miopen_logging,
|
||||
elem_id="rocm_miopen_logging",
|
||||
)
|
||||
for name, meta in rocm_vars.GENERAL_VARS.items():
|
||||
if name in rocm_vars.MIOPEN_LOGGING_VARS:
|
||||
continue
|
||||
comp = _make_component(name, meta, config)
|
||||
var_names.append(name)
|
||||
components.append(comp)
|
||||
@@ -110,6 +117,13 @@ class ROCmScript(scripts_manager.Script):
|
||||
if meta["widget"] == "dropdown" and name != "MIOPEN_GEMM_ENFORCE_BACKEND":
|
||||
comp.change(fn=lambda v, n=name: _autosave_field(n, v), inputs=[comp], outputs=[], show_progress='hidden')
|
||||
|
||||
miopen_logging_comp.change(
|
||||
fn=lambda value: rocm_mgr.set_miopen_logging(value == "Enabled"),
|
||||
inputs=[miopen_logging_comp],
|
||||
outputs=[],
|
||||
show_progress='hidden',
|
||||
)
|
||||
|
||||
_GEMM_COMPANIONS = {
|
||||
"PYTORCH_ROCM_USE_ROCBLAS": {"1": "1", "5": "0"},
|
||||
"PYTORCH_HIPBLASLT_DISABLE": {"1": "1", "5": "0"},
|
||||
@@ -141,11 +155,12 @@ class ROCmScript(scripts_manager.Script):
|
||||
gemm_comp.change(fn=gemm_changed, inputs=[gemm_comp], outputs=[style_out] + components, show_progress='hidden')
|
||||
|
||||
def apply_fn(*values):
|
||||
rocm_mgr.apply_all(var_names, list(values))
|
||||
rocm_mgr.apply_all(var_names, list(values[:-1]), miopen_logging=values[-1] == "Enabled")
|
||||
saved = rocm_mgr.load_config()
|
||||
arch = saved.get(rocm_mgr._ARCH_KEY, "")
|
||||
unavailable = rocm_profiles.UNAVAILABLE.get(arch, set())
|
||||
gemm_val = saved.get("MIOPEN_GEMM_ENFORCE_BACKEND", "1")
|
||||
logging_value = "Enabled" if saved.get("MIOPEN_DEBUG_ENABLE", "0") == "1" else "Disabled"
|
||||
result = [gr.update(value=_build_style(unavailable, gemm_val == "1"))]
|
||||
for name in var_names:
|
||||
meta = rocm_vars.ROCM_ENV_VARS[name]
|
||||
@@ -156,6 +171,7 @@ class ROCmScript(scripts_manager.Script):
|
||||
result.append(gr.update(value=rocm_mgr._dropdown_display(val, meta["options"])))
|
||||
else:
|
||||
result.append(gr.update(value=rocm_mgr._expand_venv(val)))
|
||||
result.append(gr.update(value=logging_value))
|
||||
return result
|
||||
|
||||
def reset_fn():
|
||||
@@ -164,6 +180,7 @@ class ROCmScript(scripts_manager.Script):
|
||||
arch = updated.get(rocm_mgr._ARCH_KEY, "")
|
||||
unavailable = rocm_profiles.UNAVAILABLE.get(arch, set())
|
||||
gemm_val = updated.get("MIOPEN_GEMM_ENFORCE_BACKEND", "1")
|
||||
logging_value = "Enabled" if updated.get("MIOPEN_DEBUG_ENABLE", "0") == "1" else "Disabled"
|
||||
result = [gr.update(value=_build_style(unavailable, gemm_val == "1"))]
|
||||
for name in var_names:
|
||||
meta = rocm_vars.ROCM_ENV_VARS[name]
|
||||
@@ -174,6 +191,7 @@ class ROCmScript(scripts_manager.Script):
|
||||
result.append(gr.update(value=rocm_mgr._dropdown_display(val, meta["options"])))
|
||||
else:
|
||||
result.append(gr.update(value=rocm_mgr._expand_venv(val)))
|
||||
result.append(gr.update(value=logging_value))
|
||||
return result
|
||||
|
||||
def clear_fn():
|
||||
@@ -189,6 +207,7 @@ class ROCmScript(scripts_manager.Script):
|
||||
result.append(gr.update(value=rocm_mgr._dropdown_display(meta["default"], meta["options"])))
|
||||
else:
|
||||
result.append(gr.update(value=""))
|
||||
result.append(gr.update(value="Disabled"))
|
||||
return result
|
||||
|
||||
def delete_fn():
|
||||
@@ -203,12 +222,13 @@ class ROCmScript(scripts_manager.Script):
|
||||
result.append(gr.update(value=rocm_mgr._dropdown_display(meta["default"], meta["options"])))
|
||||
else:
|
||||
result.append(gr.update(value=""))
|
||||
result.append(gr.update(value="Disabled"))
|
||||
return result
|
||||
|
||||
btn_info.click(fn=_info_html, inputs=[], outputs=[info_out], show_progress='hidden')
|
||||
btn_apply.click(fn=apply_fn, inputs=components, outputs=[style_out] + components, show_progress='hidden')
|
||||
btn_reset.click(fn=reset_fn, inputs=[], outputs=[style_out] + components, show_progress='hidden')
|
||||
btn_clear.click(fn=clear_fn, inputs=[], outputs=[style_out] + components, show_progress='hidden')
|
||||
btn_delete.click(fn=delete_fn, inputs=[], outputs=[style_out] + components, show_progress='hidden')
|
||||
btn_apply.click(fn=apply_fn, inputs=components + [miopen_logging_comp], outputs=[style_out] + components + [miopen_logging_comp], show_progress='hidden')
|
||||
btn_reset.click(fn=reset_fn, inputs=[], outputs=[style_out] + components + [miopen_logging_comp], show_progress='hidden')
|
||||
btn_clear.click(fn=clear_fn, inputs=[], outputs=[style_out] + components + [miopen_logging_comp], show_progress='hidden')
|
||||
btn_delete.click(fn=delete_fn, inputs=[], outputs=[style_out] + components + [miopen_logging_comp], show_progress='hidden')
|
||||
|
||||
return components
|
||||
return components + [miopen_logging_comp]
|
||||
|
||||
Reference in New Issue
Block a user