mirror of
https://github.com/intel/intel-graphics-compiler.git
synced 2025-10-30 08:18:26 +08:00
Add API option documentation.
Add API option documentation.
This commit is contained in:
Mielczarek, Aleksander
committed by
igcbot
igcbot
parent
bd5e98c59c
commit
363ae09a9c
268
IGC/Options/API-Options-Internal.md
Normal file
268
IGC/Options/API-Options-Internal.md
Normal file
@ -0,0 +1,268 @@
|
||||
<!--
|
||||
/*========================== begin_copyright_notice ============================
|
||||
|
||||
Copyright (C) 2025 Intel Corporation
|
||||
|
||||
SPDX-License-Identifier: MIT
|
||||
|
||||
============================= end_copyright_notice ===========================*/ -->
|
||||
# IGC API Options list
|
||||
This is an auto-generated list of options gathered from [InternalOptions.td](https://github.com/intel/intel-graphics-compiler/blob/master/IGC/Options/include/igc/Options/InternalOptions.td) file using [LLVM TableGen](https://llvm.org/docs/TableGen/).
|
||||
|
||||
The "Kind" column describes the type of Option as defined in [OptParser.td](https://github.com/llvm-mirror/llvm/blob/master/include/llvm/Option/OptParser.td#L24-L50) refer to it for further information.
|
||||
|
||||
## Table of contents
|
||||
1. [IGCInternalOption](#igcinternaloption)
|
||||
2. [VCInternalOption](#vcinternaloption)
|
||||
|
||||
## Legend
|
||||
The `cl-` and `ze-` prefixes correspond to OpenCL and Level Zero specific options. For more in-depth information regarding OpenCL options consult the OpenCL documentation[\[1\]](#references).
|
||||
|
||||
| Type | Example | Expands to |
|
||||
|:---- | :---- | :---- |
|
||||
| `<>` - Required choice | `-<cl-\|ze->option` | `-cl-option`; `-ze-option` |
|
||||
| `[]` - Optional choice | `-[cl-]option` | `-cl-option`; `-option` |
|
||||
## Option Lists
|
||||
### IGCInternalOption
|
||||
| Name | Description | Kind |
|
||||
|:---- | :---- | :----: |
|
||||
|`-D`| Manually define macros. | `KIND_JOINED_OR_SEPARATE` |
|
||||
|`-[cl-\|ze-]buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]buffer-offset-arg-required`| Tell IGC to always use buffer offset. It is valid only if -intel-has-buffer-offset-arg is present. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]compile-one-at-time`| Enables llvm::module splitting to compile only one kernel at a time. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-a64WA`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-noMaskWA`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-recompilation`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disableEUFusion`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]enable-divergent-barrier-handling`| Enable the divergent barrier pass. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exclude-ir-from-zebin`| Exclude SPIR-V section from files generated in ZEBIN format. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exp-register-file-size`| Set amount of registers used by regalloc. | `KIND_SEPARATE` |
|
||||
|`-cl-ext=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]fail-on-spill`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]force-disable-4GB-buffer`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]force-emu-int32divrem`| Use emulation fp64-based emulation functions if fp64 is supported natively. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]force-emu-sp-int32divrem`| Force the fp64-based emulation regardless of native support. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]force-enable-a64WA`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]force-global-mem-allocation`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]fp64-gen-emu`| Enable full FP64 emulation. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]functonControl`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]has-buffer-offset-arg`| This flag, together with *[-cl-intel\|-ze-opt]-greater-than-4GB-buffer-required* is used to convert stateless memory accesses, called messages or load/store, into stateful ones. The OpenCL runtime can create a surface whose base is either *buffer_base* or *buffer_base + buffer_offset*, based on whether *buffer_offset* is used. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]has-positive-pointer-offset`| For any load and store (aka message) whose address = *ptrArg + offset*, where *ptrArg* is a kernel pointer argument, offset is assumed to be non-negative if this flag is present. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]high-accuracy-nolut-math`| Enbales experimental high accuracy implementations of transcendentals. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]ignoreBFRounding`| Folds BF operands into mul/add/cmp operations. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]include-sip-csr`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]include-sip-kernel-debug`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]include-sip-kernel-local-debug`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-include-sip-csr`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]kernel-debug-enable`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]ldstcombine`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]ldstcombine_max-loadbytes`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]ldstcombine_max_storebytes`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]load-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]no-prera-scheduling`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]no-spill`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]num-thread-per-eu`| Overrides the current number of threads value defined by the user's command line option for the entire module or the compiler choice by heuristics. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->private-memory-minimal-siper-thread`| When this flag is present, it guarantees that size of private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]replace-global-offsets-by-zero`| OpenCL's global IDs are assumed to start from the origin at global offsets (offset_x, offset_y, offset_z) When this flag is present, it indicates that the global offsets are (0,0,0). | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->scratch-space-private-memory-minimal-siper-thread`| When this flag is present, it guarantees that size of scratch space private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]skip-reloc-add`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]store-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]use-32bit-ptr-arith`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-advanced-mode`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-legacy-mode`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-mode`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-printf`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]vector-coalesing`| | `KIND_SEPARATE` |
|
||||
|`-emit-visa-only`| Compile until vISA | `KIND_FLAG` |
|
||||
|`-cl-force-std`| Force a specific OpenCL C version. | `KIND_FLAG` |
|
||||
|`-m32`| | `KIND_FLAG` |
|
||||
|`-m64`| | `KIND_FLAG` |
|
||||
|`-cl-oversion=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exp-register-file-size=`| Alias for -ze-exp-register-file-size. | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-buffer-offset-arg-required`| Tell IGC to always use buffer offset. It is valid only if -intel-has-buffer-offset-arg is present. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-compile-one-at-time`| Enables llvm::module splitting to compile only one kernel at a time. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-a64WA`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-noMaskWA`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-recompilation`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disableEUFusion`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-enable-divergent-barrier-handling`| Enable the divergent barrier pass. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-exclude-ir-from-zebin`| Exclude SPIR-V section from files generated in ZEBIN format. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size`| Set amount of registers used by regalloc. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size=`| Alias for -ze-exp-register-file-size. | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-fail-on-spill`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-force-disable-4GB-buffer`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-force-emu-int32divrem`| Use emulation fp64-based emulation functions if fp64 is supported natively. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-force-emu-sp-int32divrem`| Force the fp64-based emulation regardless of native support. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-force-enable-a64WA`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-force-global-mem-allocation`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-fp64-gen-emu`| Enable full FP64 emulation. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-functonControl`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-has-buffer-offset-arg`| This flag, together with *[-cl-intel\|-ze-opt]-greater-than-4GB-buffer-required* is used to convert stateless memory accesses, called messages or load/store, into stateful ones. The OpenCL runtime can create a surface whose base is either *buffer_base* or *buffer_base + buffer_offset*, based on whether *buffer_offset* is used. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-has-positive-pointer-offset`| For any load and store (aka message) whose address = *ptrArg + offset*, where *ptrArg* is a kernel pointer argument, offset is assumed to be non-negative if this flag is present. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-high-accuracy-nolut-math`| Enbales experimental high accuracy implementations of transcendentals. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-ignoreBFRounding`| Folds BF operands into mul/add/cmp operations. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-include-sip-kernel-debug`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-include-sip-kernel-local-debug`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-kernel-debug-enable`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-ldstcombine`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-ldstcombine-max-loadbytes=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-ldstcombine-max-storebytes=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-ldstcombine=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-ldstcombine_max-loadbytes`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-ldstcombine_max_storebytes`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-load-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-load-cache-default=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-no-prera-scheduling`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-no-spill`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-num-thread-per-eu`| Overrides the current number of threads value defined by the user's command line option for the entire module or the compiler choice by heuristics. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-private-memory-minimal-siper-thread`| When this flag is present, it guarantees that size of private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-replace-global-offsets-by-zero`| OpenCL's global IDs are assumed to start from the origin at global offsets (offset_x, offset_y, offset_z) When this flag is present, it indicates that the global offsets are (0,0,0). | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-scratch-space-private-memory-minimal-siper-thread`| When this flag is present, it guarantees that size of scratch space private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-skip-reloc-add`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-store-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-store-cache-default=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-use-32bit-ptr-arith`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-advanced-mode`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-legacy-mode`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-mode`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-printf`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-vector-coalesing`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-vector-coalesing=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]ldstcombine-max-loadbytes=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]ldstcombine-max-storebytes=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]ldstcombine=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]load-cache-default=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-ze-opt-256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-ze-opt-buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-ze-opt-buffer-offset-arg-required`| Tell IGC to always use buffer offset. It is valid only if -intel-has-buffer-offset-arg is present. | `KIND_FLAG` |
|
||||
|`-ze-opt-compile-one-at-time`| Enables llvm::module splitting to compile only one kernel at a time. | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-a64WA`| | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-noMaskWA`| | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-recompilation`| | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-ze-opt-disableEUFusion`| | `KIND_FLAG` |
|
||||
|`-ze-opt-emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-ze-opt-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF. | `KIND_FLAG` |
|
||||
|`-ze-opt-enable-divergent-barrier-handling`| Enable the divergent barrier pass. | `KIND_FLAG` |
|
||||
|`-ze-opt-exclude-ir-from-zebin`| Exclude SPIR-V section from files generated in ZEBIN format. | `KIND_FLAG` |
|
||||
|`-ze-opt-exp-register-file-size`| Set amount of registers used by regalloc. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-exp-register-file-size=`| Alias for -ze-exp-register-file-size. | `KIND_JOINED` |
|
||||
|`-ze-opt-fail-on-spill`| | `KIND_FLAG` |
|
||||
|`-ze-opt-force-disable-4GB-buffer`| | `KIND_FLAG` |
|
||||
|`-ze-opt-force-emu-int32divrem`| Use emulation fp64-based emulation functions if fp64 is supported natively. | `KIND_FLAG` |
|
||||
|`-ze-opt-force-emu-sp-int32divrem`| Force the fp64-based emulation regardless of native support. | `KIND_FLAG` |
|
||||
|`-ze-opt-force-enable-a64WA`| | `KIND_FLAG` |
|
||||
|`-ze-opt-force-global-mem-allocation`| | `KIND_FLAG` |
|
||||
|`-ze-opt-fp64-gen-emu`| Enable full FP64 emulation. | `KIND_FLAG` |
|
||||
|`-ze-opt-functonControl`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-ze-opt-has-buffer-offset-arg`| This flag, together with *[-cl-intel\|-ze-opt]-greater-than-4GB-buffer-required* is used to convert stateless memory accesses, called messages or load/store, into stateful ones. The OpenCL runtime can create a surface whose base is either *buffer_base* or *buffer_base + buffer_offset*, based on whether *buffer_offset* is used. | `KIND_FLAG` |
|
||||
|`-ze-opt-has-positive-pointer-offset`| For any load and store (aka message) whose address = *ptrArg + offset*, where *ptrArg* is a kernel pointer argument, offset is assumed to be non-negative if this flag is present. | `KIND_FLAG` |
|
||||
|`-ze-opt-high-accuracy-nolut-math`| Enbales experimental high accuracy implementations of transcendentals. | `KIND_FLAG` |
|
||||
|`-ze-opt-ignoreBFRounding`| Folds BF operands into mul/add/cmp operations. | `KIND_FLAG` |
|
||||
|`-ze-opt-include-sip-csr`| | `KIND_FLAG` |
|
||||
|`-ze-opt-include-sip-kernel-debug`| | `KIND_FLAG` |
|
||||
|`-ze-opt-include-sip-kernel-local-debug`| | `KIND_FLAG` |
|
||||
|`-ze-opt-kernel-debug-enable`| | `KIND_FLAG` |
|
||||
|`-ze-opt-ldstcombine`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-ldstcombine-max-loadbytes=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-ldstcombine-max-storebytes=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-ldstcombine=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-ldstcombine_max-loadbytes`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-ldstcombine_max_storebytes`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-load-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-load-cache-default=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-ze-opt-no-prera-scheduling`| | `KIND_FLAG` |
|
||||
|`-ze-opt-no-spill`| | `KIND_FLAG` |
|
||||
|`-ze-opt-num-thread-per-eu`| Overrides the current number of threads value defined by the user's command line option for the entire module or the compiler choice by heuristics. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-private-memory-minimal-size-per-thread`| When this flag is present, it guarantees that size of private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-replace-global-offsets-by-zero`| OpenCL's global IDs are assumed to start from the origin at global offsets (offset_x, offset_y, offset_z) When this flag is present, it indicates that the global offsets are (0,0,0). | `KIND_FLAG` |
|
||||
|`-ze-opt-scratch-space-private-memory-minimal-size-per-thread`| When this flag is present, it guarantees that size of scratch space private memory allocated per thread can not be less then the given value. Constraint: <SIZE> >= 0. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-skip-reloc-add`| | `KIND_FLAG` |
|
||||
|`-ze-opt-store-cache-default`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-store-cache-default=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-use-32bit-ptr-arith`| | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-advanced-mode`| | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-legacy-mode`| | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-mode`| | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-printf`| | `KIND_FLAG` |
|
||||
|`-ze-opt-vector-coalesing`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-vector-coalesing=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]store-cache-default=`| | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]vector-coalesing=`| | `KIND_JOINED` |
|
||||
|
||||
### VCInternalOption
|
||||
| Name | Description | Kind |
|
||||
|:---- | :---- | :----: |
|
||||
|`-binary-format`| Set in which format should be generated binary; values: 'cm' or 'ze' | `KIND_SEPARATE` |
|
||||
|`-binary-format=`| Alias for -binary-format <value> | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
|`-dump-asm`| Dump assembly (visaasm, asm, etc.) | `KIND_FLAG` |
|
||||
|`-dump-isa-binary`| Dump isa binary after finalization pass | `KIND_FLAG` |
|
||||
|`-dump-llvm-ir`| Dump llvm IR after SPIRV reading, optimizations and codegen | `KIND_FLAG` |
|
||||
|`-emit-visa-only`| Compile until vISA | `KIND_FLAG` |
|
||||
|`-flush-l3-for-global`| Enable flushing L3 cache for globals | `KIND_FLAG` |
|
||||
|`-freset-llvm-stats`| Reset performance metrics before compilation | `KIND_FLAG` |
|
||||
|`-freset-time-report`| Reset timing summary before compilation | `KIND_FLAG` |
|
||||
|`-ftime-report`| Print timing summary of each stage of compilation | `KIND_FLAG` |
|
||||
|`-gpu-scope-fence-on-single-tile`| Allow the use of | `KIND_FLAG` |
|
||||
|`-help`| Display available API options | `KIND_FLAG` |
|
||||
|`-help-internal`| Display available internal options | `KIND_FLAG` |
|
||||
|`-llvm-options`| Additional options forwarded to llvm CommandLine global option parser | `KIND_SEPARATE` |
|
||||
|`-llvm-options=`| Alias for -llvm-options | `KIND_JOINED` |
|
||||
|`-mdump_asm`| Alias for -dump-asm | `KIND_FLAG` |
|
||||
|`-print-stats`| Print performance metrics and statistics | `KIND_FLAG` |
|
||||
|`-runtime`| Set runtime for which binary should be generated; values: 'cm' or 'ze' | `KIND_SEPARATE` |
|
||||
|`-runtime=`| Alias for -runtime <value> | `KIND_JOINED` |
|
||||
|`-stats-file`| Filename to write statistics to | `KIND_SEPARATE` |
|
||||
|`-stats-file=`| Alias for -stats-file | `KIND_JOINED` |
|
||||
|`-target-features`| Auxiliary target features | `KIND_SEPARATE` |
|
||||
|`-target-features=`| Alias for -target-features | `KIND_JOINED` |
|
||||
|`-vc-disable-debuggable-kernels`| Disable emission of debuggable kernels for legacy path | `KIND_FLAG` |
|
||||
|`-vc-ignore-loop-unroll-threshold-on-pragma`| Ignore loop unroll threshold on pragma for VC | `KIND_FLAG` |
|
||||
|`-vc-interop-subgroup-size`| Set subgroup size used for cross-module calls/returns | `KIND_SEPARATE` |
|
||||
|`-vc-loop-unroll-threshold`| Loop unroll threshold for VC | `KIND_SEPARATE` |
|
||||
|`-vc-loop-unroll-threshold=`| Alias for -vc-loop-unroll-threshold <value> | `KIND_JOINED` |
|
||||
|`-vc-report-lsc-stores-with-non-default-l1-cache-controls`| Ignore loop unroll threshold on pragma for VC | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
|`-ze-opt-buffer-bounds-checking`| Enable buffer bounds checking | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-sendwarwa`| Disable SendWAR WA PVC platform only | `KIND_FLAG` |
|
||||
|`-ze-opt-emit-zebin-visa-sections`| Add vISA asm as sections in ZeBin | `KIND_FLAG` |
|
||||
|`-ze-opt-minimum-valid-address-checking`| Set minimal valid address | `KIND_SEPARATE` |
|
||||
|`-ze-opt-use-bindless-buffers`| Use bindless mode for buffers | `KIND_FLAG` |
|
||||
|`-ze-opt-use-bindless-images`| Use bindless mode for images | `KIND_FLAG` |
|
||||
### References
|
||||
[1]: [OpenCL documentation](https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#compiler-options)
|
||||
262
IGC/Options/API-Options.md
Normal file
262
IGC/Options/API-Options.md
Normal file
@ -0,0 +1,262 @@
|
||||
<!--
|
||||
/*========================== begin_copyright_notice ============================
|
||||
|
||||
Copyright (C) 2025 Intel Corporation
|
||||
|
||||
SPDX-License-Identifier: MIT
|
||||
|
||||
============================= end_copyright_notice ===========================*/ -->
|
||||
# IGC API Options list
|
||||
This is an auto-generated list of options gathered from [ApiOptions.td](https://github.com/intel/intel-graphics-compiler/blob/master/IGC/Options/include/igc/Options/ApiOptions.td) file using [LLVM TableGen](https://llvm.org/docs/TableGen/).
|
||||
|
||||
The "Kind" column describes the type of Option as defined in [OptParser.td](https://github.com/llvm-mirror/llvm/blob/master/include/llvm/Option/OptParser.td#L24-L50) refer to it for further information.
|
||||
|
||||
## Table of contents
|
||||
- [IGC API Options list](#igc-api-options-list)
|
||||
- [Table of contents](#table-of-contents)
|
||||
- [Legend](#legend)
|
||||
- [Option Lists](#option-lists)
|
||||
- [IGCApiOption](#igcapioption)
|
||||
- [IgcmcApiOption](#igcmcapioption)
|
||||
- [VCApiOption](#vcapioption)
|
||||
- [References](#references)
|
||||
|
||||
## Legend
|
||||
The `cl-` and `ze-` prefixes correspond to OpenCL and Level Zero specific options. For more in-depth information regarding OpenCL options consult the OpenCL documentation[\[1\]](#references).
|
||||
|
||||
| Type | Example | Expands to |
|
||||
|:---- | :---- | :---- |
|
||||
| `<>` - Required choice | `-<cl-\|ze->option` | `-cl-option`; `-ze-option` |
|
||||
| `[]` - Optional choice | `-[cl-]option` | `-cl-option`; `-option` |
|
||||
## Option Lists
|
||||
### IGCApiOption
|
||||
| Name | Description | Kind |
|
||||
|:---- | :---- | :----: |
|
||||
|`-D`| Manually define macros. | `KIND_JOINED_OR_SEPARATE` |
|
||||
|`-g`| Enable generation of debug information and enables kernel debug | `KIND_FLAG` |
|
||||
|`-I`| Add directory to the list of directories which will be searched for header files. | `KIND_JOINED_OR_SEPARATE` |
|
||||
|`-s`| Strip all symbol table and debug informaton from the output binary. | `KIND_SEPARATE` |
|
||||
|`-w`| Remove all warning messages. | `KIND_FLAG` |
|
||||
|`-x`| Manualy provide type of file. Takes only spir or spir64 as argument. | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-cl-denorms-are-zero`| Controls how single precision and double precision denormalized numbers are handled. If specified as a build option, the single precision denormalized numbers may be flushed to zero; double precision denormalized numbers may also be flushed to zero if the optional extension for double precision is supported | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]emit-lib-compile-errors`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]enable-ieee-float-exception-trap`| This flag enables the IEEE exception trap bit in Control Register in the prolog of the kernel. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-cl-fast-relaxed-math`| Sets the optimization options -cl-finite-math-only and -cl-unsafe-math-optimizations. This option causes the preprocessor macro __FAST_RELAXED_MATH__ to be defined in the OpenCL program. | `KIND_FLAG` |
|
||||
|`-cl-finite-math-only`| Allow optimizations for floating-point arithmetic that assume that arguments and results are not NaNs, +Inf, -Inf. This option may violate the OpenCL numerical compliance requirements for single precision and double precision floating-point, as well as edge case behavior. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]fp32-correctly-rounded-divide-sqrt`| Allows an application to specify that single precision floating-point divide (x/y and 1/x) and sqrt used in the program source are correctly rounded. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]fp64-gen-conv-emu`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]fp64-gen-emu`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-cl-intel-debug-info`| | `KIND_FLAG` |
|
||||
|`-cl-intel-disable-a64WA`| | `KIND_FLAG` |
|
||||
|`-cl-kernel-arg-info`| Allow the compiler to store information about the arguments of a kernel(s) in the program executable. The argument information stored includes the argument name, its type, the address space and access qualifiers used. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]large-grf-kernel`| -ze-opt-large-grf-kernel <string> tells IGC to use large GRF size if kernel name contains <string> regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->library-compilation`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-cl-mad-enable`| Allow a * b + c to be replaced by a mad instruction. | `KIND_FLAG` |
|
||||
|`-cl-match-sincospi`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]no-local-to-generic`| | `KIND_FLAG` |
|
||||
|`-cl-no-signed-zeros`| Allow optimizations for floating-point arithmetic that ignore the signedness of zero. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]no-subgroup-ifp`| This indicates that kernels in this program do not require sub-groups to make independent forward progress. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]poison-unsupported-fp64-kernels`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]regular-grf-kernel`| -ze-opt-regular-grf-kernel <string> tells IGC to use regular GRF size if kernel name contains <string >regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]reqd-eu-thread-count`| | `KIND_SEPARATE` |
|
||||
|`-cl-single-precision-constant`| Forces implicit conversions of double-precision floating-point literals to single precision. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]skip-fde`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]static-profile-guided-trimming`| Enable static analysis in the kernel trimming. | `KIND_FLAG` |
|
||||
|`-cl-std=`| Determine the language version to be accepted by the compiler. | `KIND_JOINED` |
|
||||
|`-cl-strict-aliasing`| Allows the compiler to assume the strictest aliasing rules. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]take-global-address`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]uniform-work-group-size`| This requires that the global work-size be a multiple of the work-group size specified to clEnqueueNDRangeKernel. | `KIND_FLAG` |
|
||||
|`-cl-unsafe-math-optimizations`| Allow optimizations for floating-point arithmetic that (a) assume that arguments and results are valid, (b) may violate the IEEE 754 standard, (c) assume relaxed OpenCL numerical compliance requirements as defined in the unsafe math optimization section of the OpenCL C or OpenCL SPIR-V Environment specifications, and (d) may violate edge case behavior in the OpenCL C or OpenCL SPIR-V Environment specifications. | `KIND_FLAG` |
|
||||
|`-dump-opt-llvm=`| Dump the llvm output to the specified file. | `KIND_JOINED` |
|
||||
|`-dwarf-column-info`| | `KIND_FLAG` |
|
||||
|`-gline-tables-only`| Generate only line table debug information. | `KIND_FLAG` |
|
||||
|`-igc_opts`| Pass IGC options delimited by ',' or ' '. | `KIND_JOINED_OR_SEPARATE` |
|
||||
|`-profiler`| | `KIND_FLAG` |
|
||||
|`-spir-std=`| Specify the SPIR version. | `KIND_JOINED` |
|
||||
|`-triple`| | `KIND_SEPARATE` |
|
||||
|`-Werror`| Treat every warning as an error. | `KIND_FLAG` |
|
||||
|`-Xfinalizer`| Pass <arg> to the visa finalizer | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-emit-lib-compile-errors`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-enable-ieee-float-exception-trap`| This flag enables the IEEE exception trap bit in Control Register in the prolog of the kernel. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-fp32-correctly-rounded-divide-sqrt`| Allows an application to specify that single precision floating-point divide (x/y and 1/x) and sqrt used in the program source are correctly rounded. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-fp64-gen-conv-emu`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-fp64-gen-emu`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-large-grf-kernel`| -ze-opt-large-grf-kernel <string> tells IGC to use large GRF size if kernel name contains <string> regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-library-compilation`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-no-local-to-generic`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-no-subgroup-ifp`| This indicates that kernels in this program do not require sub-groups to make independent forward progress. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-poison-unsupported-fp64-kernels`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-regular-grf-kernel`| -ze-opt-regular-grf-kernel <string> tells IGC to use regular GRF size if kernel name contains <string >regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-reqd-eu-thread-count`| | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-reqd-eu-thread-count=`| | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-skip-fde`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-static-profile-guided-trimming`| Enable static analysis in the kernel trimming. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-take-global-address`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-uniform-work-group-size`| This requires that the global work-size be a multiple of the work-group size specified to clEnqueueNDRangeKernel. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-ze-opt-128-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-ze-opt-256-GRF-per-thread`| | `KIND_FLAG` |
|
||||
|`-ze-opt-collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-ze-opt-emit-lib-compile-errors`| | `KIND_FLAG` |
|
||||
|`-ze-opt-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-ze-opt-enable-ieee-float-exception-trap`| This flag enables the IEEE exception trap bit in Control Register in the prolog of the kernel. | `KIND_FLAG` |
|
||||
|`-ze-opt-exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-ze-opt-exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-ze-opt-fp32-correctly-rounded-divide-sqrt`| Allows an application to specify that single precision floating-point divide (x/y and 1/x) and sqrt used in the program source are correctly rounded. | `KIND_FLAG` |
|
||||
|`-ze-opt-fp64-gen-conv-emu`| | `KIND_FLAG` |
|
||||
|`-ze-opt-fp64-gen-emu`| | `KIND_FLAG` |
|
||||
|`-ze-opt-greater-than-4GB-buffer-required`| When this flag is present, it indicates that any OpenCL buffers can be more than 4GB in size. If it is absent, all buffers are not more than 4GB in size. | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-ze-opt-gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-ze-opt-large-grf-kernel`| -ze-opt-large-grf-kernel <string> tells IGC to use large GRF size if kernel name contains <string> regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-ze-opt-library-compilation`| | `KIND_FLAG` |
|
||||
|`-ze-opt-library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-ze-opt-library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-ze-opt-no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-ze-opt-no-local-to-generic`| | `KIND_FLAG` |
|
||||
|`-ze-opt-no-subgroup-ifp`| This indicates that kernels in this program do not require sub-groups to make independent forward progress. | `KIND_FLAG` |
|
||||
|`-ze-opt-opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-ze-opt-poison-unsupported-fp64-kernels`| | `KIND_FLAG` |
|
||||
|`-ze-opt-regular-grf-kernel`| -ze-opt-regular-grf-kernel <string> tells IGC to use regular GRF size if kernel name contains <string >regardless of module-level options. | `KIND_SEPARATE` |
|
||||
|`-ze-opt-reqd-eu-thread-count`| | `KIND_SEPARATE` |
|
||||
|`-ze-opt-reqd-eu-thread-count=`| | `KIND_JOINED` |
|
||||
|`-ze-opt-skip-fde`| | `KIND_FLAG` |
|
||||
|`-ze-opt-static-profile-guided-trimming`| Enable static analysis in the kernel trimming. | `KIND_FLAG` |
|
||||
|`-ze-opt-take-global-address`| | `KIND_FLAG` |
|
||||
|`-ze-opt-uniform-work-group-size`| This requires that the global work-size be a multiple of the work-group size specified to clEnqueueNDRangeKernel. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]reqd-eu-thread-count=`| | `KIND_JOINED` |
|
||||
|
||||
### IgcmcApiOption
|
||||
| Name | Description | Kind |
|
||||
|:---- | :---- | :----: |
|
||||
|`-cmc`| Enable igcmc compatible mode; incompatible with -vc-codegen | `KIND_FLAG` |
|
||||
|`-doubleGRF`| Alias for -ze-opt-large-register-file | `KIND_FLAG` |
|
||||
|`-no_vector_decomposition`| Alias for -ze-no-vector-decomposition | `KIND_FLAG` |
|
||||
|`-stack-mem-size=`| Control stack memory size (in bytes) | `KIND_JOINED` |
|
||||
|`-visaopts=`| Options for finalizer in form | `KIND_JOINED` |
|
||||
|
||||
### VCApiOption
|
||||
| Name | Description | Kind |
|
||||
|:---- | :---- | :----: |
|
||||
|`-g`| Enable generation of debug information and enables kernel debug | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->library-compilation`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]skip-fde`| | `KIND_FLAG` |
|
||||
|`-codegen-optimize`| Set codegen optimization level to either 'none' or 'full' | `KIND_SEPARATE` |
|
||||
|`-codegen-optimize=`| Alias for -codegen-optimize | `KIND_JOINED` |
|
||||
|`-depressurizer-flag-grf-tolerance`| Threshold for disabling flag pressure reduction | `KIND_SEPARATE` |
|
||||
|`-depressurizer-flag-grf-tolerance=`| Alias for -depressurizer-flag-grf-tolerance <value> | `KIND_JOINED` |
|
||||
|`-depressurizer-grf-threshold`| Threshold for GRF pressure reduction | `KIND_SEPARATE` |
|
||||
|`-depressurizer-grf-threshold=`| Alias for -depressurizer-grf-threshold <value> | `KIND_JOINED` |
|
||||
|`-disable-finalizer-msg`| Turns off critical messages from cISA Builder | `KIND_FLAG` |
|
||||
|`-double-GRF`| Alias for -ze-opt-large-register-file | `KIND_FLAG` |
|
||||
|`-doubleGRF`| Alias for -ze-opt-large-register-file | `KIND_FLAG` |
|
||||
|`-ffp-contract=`| Form fused FP ops: fast (everywhere) \| on (nowhere except insertion fma) \| off (never fuse) | `KIND_JOINED` |
|
||||
|`-fno-jump-tables`| Disable the use of jump tables for lowering switches | `KIND_FLAG` |
|
||||
|`-fno-struct-splitting`| Disable StructSplitter pass | `KIND_FLAG` |
|
||||
|`-ftranslate-legacy-memory-intrinsics`| Legalize old memory intrinsics | `KIND_FLAG` |
|
||||
|`-no-optimize`| Alias for -optimize=none | `KIND_FLAG` |
|
||||
|`-[ze-]no-vector-decomposition`| Alias for -ze-no-vector-decomposition | `KIND_FLAG` |
|
||||
|`-no_vector_decomposition`| Alias for -ze-no-vector-decomposition | `KIND_FLAG` |
|
||||
|`-optimize`| Set optimization level to either 'none' or 'full' | `KIND_SEPARATE` |
|
||||
|`-optimize=`| Alias for -optimize | `KIND_JOINED` |
|
||||
|`-stateless-stack-mem-size=`| Stateless memory amount per thread | `KIND_JOINED` |
|
||||
|`-vc-codegen`| Enable vector codegenerator | `KIND_FLAG` |
|
||||
|`-vc-disable-indvars-opt`| Disable induction variable optimization | `KIND_FLAG` |
|
||||
|`-vc-disable-non-overlapping-region-opt`| Disable non-overlapping region optimization | `KIND_FLAG` |
|
||||
|`-vc-enable-preemption`| Enable preemption for VC path | `KIND_FLAG` |
|
||||
|`-vc-use-bindless-buffers`| Enable bindless buffer access | `KIND_FLAG` |
|
||||
|`-vc-use-bindless-images`| Enable bindless image access | `KIND_FLAG` |
|
||||
|`-vc-use-plain-2d-images`| Treat image2d_t annotation as non-media image | `KIND_FLAG` |
|
||||
|`-Xfinalizer`| Pass <arg> to the visa finalizer | `KIND_SEPARATE` |
|
||||
|`-[cl-\|ze-]exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-[cl-\|ze-]gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-library-compilation`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-<cl-\|ze->intel-library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-<cl-\|ze->intel-no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-<cl-\|ze->intel-skip-fde`| | `KIND_FLAG` |
|
||||
|`-ze-kernel-debug-enable`| Alias for -g | `KIND_FLAG` |
|
||||
|`-[cl-\|ze-]library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-ze-opt-collect-cost-info`| Find argument symbols, calculate loop count and loop cost expressions | `KIND_FLAG` |
|
||||
|`-ze-opt-disable-compaction`| Disable compaction pass in finalizer. This pass is used to decide whether to use a compacted, i.e. shorter encoding of machine instructions wherever ISA allows. | `KIND_FLAG` |
|
||||
|`-ze-opt-enable-auto-large-GRF-mode`| Use compiler heuristics to determine number of GRF | `KIND_FLAG` |
|
||||
|`-ze-opt-exp-register-file-size`| Set amount of registers used by regalloc | `KIND_SEPARATE` |
|
||||
|`-ze-opt-exp-register-file-size=`| Alias for -ze-exp-register-file-size | `KIND_JOINED` |
|
||||
|`-ze-opt-gtpin-grf-info`| Enable vISA grf-info interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-indir-ref`| Ask finalizer to provide list of registers used by indirect operand per %ip | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-rera`| Enables vISA re_ra interface for gtpin structure generation | `KIND_FLAG` |
|
||||
|`-ze-opt-gtpin-scratch-area-size`| Set gtpin scratch area size | `KIND_SEPARATE` |
|
||||
|`-ze-opt-gtpin-scratch-area-size=`| Alias for -ze-gtpin-scratch-area-size | `KIND_JOINED` |
|
||||
|`-ze-opt-large-register-file`| Increase number of available GRF | `KIND_FLAG` |
|
||||
|`-ze-opt-library-compilation`| | `KIND_FLAG` |
|
||||
|`-ze-opt-library-compile-simd`| Select SIMD size for library compilations [8\|16\|32] | `KIND_SEPARATE` |
|
||||
|`-ze-opt-library-compile-simd=`| Select SIMD size for library compilations [8\|16\|32] | `KIND_JOINED` |
|
||||
|`-ze-opt-no-fusedCallWA`| | `KIND_FLAG` |
|
||||
|`-ze-opt-opt-disable`| Turns off optimizations | `KIND_FLAG` |
|
||||
|`-ze-opt-skip-fde`| | `KIND_FLAG` |
|
||||
### References
|
||||
[1]: [OpenCL documentation](https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#compiler-options)
|
||||
Reference in New Issue
Block a user