Home
last modified time | relevance | path

Searched full:waves (Results 1 – 25 of 30) sorted by relevance

12

/freebsd/contrib/llvm-project/llvm/lib/Target/AMDGPU/
H A DAMDGPUSubtarget.h92 /// \returns Subtarget's default pair of minimum/maximum number of waves per
93 /// execution unit for function \p F, or minimum/maximum number of waves per
94 /// execution unit explicitly requested using "amdgpu-waves-per-eu" attribute
99 /// compatible with minimum/maximum number of waves limited by flat work group
274 /// \returns Number of waves per execution unit required to support the given
279 /// \returns Minimum number of waves per execution unit supported by the
283 /// \returns Maximum number of waves per execution unit supported by the
H A DGCNSubtarget.h1306 /// Return the maximum number of waves per SIMD for kernels using \p SGPRs
1310 /// Return the maximum number of waves per SIMD for kernels using \p VGPRs
1405 /// \returns Minimum number of SGPRs that meets the given number of waves per
1411 /// \returns Maximum number of SGPRs that meets the given number of waves per
1435 /// \returns Maximum number of SGPRs that meets number of waves per execution
1439 /// \returns Value that meets number of waves per execution unit requirement
1441 /// subtarget's specifications, or does not meet number of waves per execution
1445 /// \returns Maximum number of SGPRs that meets number of waves per execution
1449 /// \returns Value that meets number of waves per execution unit requirement
1451 /// subtarget's specifications, or does not meet number of waves per execution
[all …]
H A DSIMemoryLegalizer.cpp1132 // not needed as LDS operations for all waves are executed in a total in insertWait()
1133 // global ordering as observed by all waves. Required if also in insertWait()
1154 // is not needed as GDS operations for all waves are executed in a total in insertWait()
1155 // global ordering as observed by all waves. Required if also in insertWait()
1310 // In threadgroup split mode the waves of a work-group can be executing on in enableLoadCacheBypass()
1312 // Otherwise in non-threadgroup split mode all waves of a work-group are in enableLoadCacheBypass()
1451 // In threadgroup split mode the waves of a work-group can be executing on in insertWait()
1453 // to complete to ensure they are visible to waves in the other CUs. in insertWait()
1454 // Otherwise in non-threadgroup split mode all waves of a work-group are on in insertWait()
1455 // the same CU, so no need to wait for global memory as all waves in the in insertWait()
[all …]
H A DAMDGPUSubtarget.cpp342 // allows the given function to achieve an occupancy of NWaves waves per
360 // Returns the maximum occupancy, in number of waves per SIMD / EU, that can
385 // Round to the number of waves per CU. in getOccupancyWithLocalMemSize()
389 // Number of waves per EU (SIMD). in getOccupancyWithLocalMemSize()
392 // Clamp to the maximum possible number of waves. in getOccupancyWithLocalMemSize()
450 // Default minimum/maximum number of waves per execution unit. in getEffectiveWavesPerEU()
455 // number of waves per execution unit to values implied by requested in getEffectiveWavesPerEU()
480 // Default minimum/maximum number of waves per execution unit. in getWavesPerEU()
483 // Requested minimum/maximum number of waves per execution unit. in getWavesPerEU()
485 AMDGPU::getIntegerPairAttribute(F, "amdgpu-waves-per-eu", Default, true); in getWavesPerEU()
[all …]
H A DSIProgramInfo.h76 // Number of SGPRs that meets number of waves per execution unit request.
79 // Number of VGPRs that meets number of waves per execution unit request.
H A DR600Subtarget.h151 /// \returns Number of waves per execution unit required to support the given
158 /// \returns Minimum number of waves per execution unit supported by the
H A DSIRegisterInfo.h410 /// Return all SGPR128 which satisfy the waves per execution unit requirement
414 /// Return all SGPR64 which satisfy the waves per execution unit requirement
418 /// Return all SGPR32 which satisfy the waves per execution unit requirement
H A DAMDGPUPerfHintAnalysis.h11 /// kernel may benefit from limiting number of waves to reduce cache thrashing.
H A DSIMachineFunctionInfo.h423 // A pair of default/requested minimum/maximum number of waves per execution
1038 /// \returns A pair of default/requested minimum/maximum number of waves per
1044 /// \returns Default/requested minimum number of waves per execution unit.
1049 /// \returns Default/requested maximum number of waves per execution unit.
H A DAMDGPUMachineFunction.h67 // Kernel may need limited waves per EU for better performance.
H A DAMDGPUSetWavePriority.cpp12 /// waves to issue their VMEM instructions as well.
H A DAMDGPUAttributor.cpp181 /// Get the effective value of "amdgpu-waves-per-eu" for the function,
818 /// Propagate amdgpu-waves-per-eu attribute.
821 : AAAMDSizeRangeAttribute(IRP, A, "amdgpu-waves-per-eu") {} in AAAMDWavesPerEU()
H A DAMDGPUAsmPrinter.cpp908 // number of waves per execution unit request. in getSIProgramInfo()
1088 AMDGPU::getIntegerPairAttribute(F, "amdgpu-waves-per-eu", {0, 0}, true); in getSIProgramInfo()
1093 "failed to meet occupancy target given by 'amdgpu-waves-per-eu' in " in getSIProgramInfo()
1534 EmitResourceUsageRemark("Occupancy", "Occupancy [waves/SIMD]", in emitResourceUsageRemarks()
H A DGCNSchedStrategy.h122 /// maximum number of waves per simd).
H A DAMDGPUPerfHintAnalysis.cpp11 /// kernel may benefit from limiting number of waves to reduce cache thrashing.
H A DGCNSchedStrategy.cpp1047 // Allow memory bound functions to drop to 4 waves if not limited by an in checkScheduling()
1052 << MFI.getMinAllowedOccupancy() << " waves\n"); in checkScheduling()
H A DAMDKernelCodeT.h333 /// apply to all waves of the grid. It is possible to specify more than 16 User
/freebsd/contrib/llvm-project/llvm/lib/Target/AMDGPU/Utils/
H A DAMDGPUBaseInfo.h217 /// \returns Minimum number of waves per execution unit for given subtarget \p
221 /// \returns Maximum number of waves per execution unit for given subtarget \p
225 /// \returns Number of waves per execution unit required to support the given \p
236 /// \returns Number of waves per work group for given subtarget \p STI and
253 /// \returns Minimum number of SGPRs that meets the given number of waves per
257 /// \returns Maximum number of SGPRs that meets the given number of waves per
304 /// \returns Minimum number of VGPRs that meets given number of waves per
308 /// \returns Maximum number of VGPRs that meets given number of waves per
312 /// \returns Number of waves reachable for a given \p NumVGPRs usage for given
317 /// \returns Number of waves reachable for a given \p NumVGPRs usage, \p Granule
H A DAMDGPUBaseInfo.cpp915 // "Per CU" really means "per whatever functional block the waves of a in getLocalMemorySize()
933 // "Per CU" really means "per whatever functional block the waves of a in getEUsPerCU()
/freebsd/sys/contrib/device-tree/Bindings/iio/proximity/
H A Ddevantech-srf04.yaml48 Thus this GPIO is set while the ultrasonic waves are doing one round
/freebsd/contrib/llvm-project/clang/lib/CodeGen/Targets/
H A DAMDGPU.cpp686 F->addFnAttr("amdgpu-waves-per-eu", AttrVal); in handleAMDGPUWavesPerEUAttr()
/freebsd/contrib/file/magic/Magdir/
H A Dgames320 >4 ulelong 0x0000032d (Waves)
/freebsd/contrib/llvm-project/llvm/include/llvm/IR/
H A DIntrinsicsAMDGPU.td2129 // bar_val is the total number of waves that will wait on this
2141 // bar_val is the total number of waves that will wait on this
/freebsd/contrib/llvm-project/clang/lib/CodeGen/
H A DCodeGenModule.h1650 /// Emit the IR encoding to attach the AMD GPU waves-per-eu attribute to \p F.
/freebsd/contrib/llvm-project/clang/include/clang/Basic/
H A DAttrDocs.td2702 ``<min>`` parameter specifies the requested minimum number of waves per EU, and
2703 *optional* ``<max>`` parameter specifies the requested maximum number of waves
2705 then there is no restriction on the maximum number of waves per EU other than

12