Home
last modified time | relevance | path

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

12

/llvm-project/mlir/include/mlir-c/Dialect/
H A DQuant.h28 /// Returns the bit flag used to indicate signedness of a quantized type.
31 /// Returns the minimum possible value stored by a quantized type.
35 /// Returns the maximum possible value stored by a quantized type.
39 /// Gets the original type approximated by the given quantized type.
42 /// Gets the flags associated with the given quantized type.
51 /// Returns the minimum value that the storage type of the given quantized type
55 /// Returns the maximum value that the storage type of the given quantized type
59 /// Returns the integral bitwidth that the storage type of the given quantized
65 /// quantized `type`.
69 /// Returns the element type of the given quantized type as another quantized
[all …]
/llvm-project/mlir/include/mlir/Dialect/Quant/
H A DQuantOps.td
H A DQuantTypes.h
H A DQuantOpsBase.td
H A DFakeQuantSupport.h
H A DUniformSupport.h
/llvm-project/mlir/lib/Bindings/Python/
H A DDialectQuant.cpp50 "Type expressed by this quantized type."); in populateDialectQuantSubmodule()
53 "Flags of this quantized type (named accessors should be preferred to " in populateDialectQuantSubmodule()
58 "Signedness of this quantized type."); in populateDialectQuantSubmodule()
62 "Storage type backing this quantized type."); in populateDialectQuantSubmodule()
66 "The minimum value held by the storage type of this quantized type."); in populateDialectQuantSubmodule()
70 "The maximum value held by the storage type of this quantized type."); in populateDialectQuantSubmodule()
76 "The bitwidth of the storage type of this quantized type."); in populateDialectQuantSubmodule()
82 "Checks whether the candidate type can be expressed by this quantized " in populateDialectQuantSubmodule()
90 "Element type of this quantized type expressed as quantized typ in populateDialectQuantSubmodule()
[all...]
/llvm-project/mlir/lib/Dialect/Tosa/Utils/
H A DQuantUtils.cpp94 /// Generates a quantized multiplier/shift from double.
134 // Weights must be either per-tensor quantized or per-axis quantized. in buildConvOpQuantizationAttr()
136 "Weights must be either per-tensor or per-axis quantized"); in buildConvOpQuantizationAttr()
138 // Either all quantized or all not quantized. in buildConvOpQuantizationAttr()
141 "Inputs and weights must be all quantized or all not quantized"); in buildConvOpQuantizationAttr()
176 // A and B are either all quantized or all not quantized. in buildMatMulOpQuantizationAttr()
178 "Matmul operands must be all quantized or all not quantized"); in buildMatMulOpQuantizationAttr()
205 // Either all quantized or all not quantized. in buildUnaryOpQuantizationAttr()
207 "Unary inputs/outputs must be all quantized or all not quantized"); in buildUnaryOpQuantizationAttr()
237 /// Builds output type for a quantized ConvOp with the right bitwidth.
[all …]
/llvm-project/mlir/test/lib/Dialect/Tosa/
H A DTosaTestPasses.cpp28 // This transformation converts quantized uint8 to quantized int8. The
56 // skip if output is not per-tensor quantized type. in matchAndRewrite()
98 // This transformation modifies the quantized output of a test conv2d input and
141 // Works on quantized type only. in matchAndRewrite()
154 // Create rescale to quantized type in matchAndRewrite()
165 // Obtain the quantized scale = multiplier and shift. in matchAndRewrite()
/llvm-project/mlir/docs/
H A DQuantization.md188 * A family of [QuantizedTypes](#quantized-type) which represent the
192 * [Type conversion ops](#quantized-type-conversion-operations) for converting
203 * The TFLite op-set natively supports uniform-quantized variants.
205 to the TFLite quantized operation set.
213 ### Quantized type
220 ### Quantized type conversion operations
245 operations at inference time. When quantized inference was enabled, if every
249 judgment about how to convert to use kernels from its quantized operations subset.
260 This scheme also naturally allows computations that are *partially quantized*
/llvm-project/mlir/test/python/dialects/
H A Dquant.py62 # CHECK: quantized element type: !quant.any<i8<-8:7>:f32>
63 print(f"quantized element type: {any.quantized_element_type}")
115 # CHECK: quantized dim: 1
116 print(f"quantized dim: {per_axis.quantized_dimension}")
/llvm-project/mlir/test/Dialect/Quant/
H A Dparse-uniform-invalid.mlir5 // expected-error@+1 {{unknown quantized type foobar}}
10 // expected-error@+1 {{unknown quantized type __}}
114 // Illegal uniform params: missing quantized dimension
119 // Illegal uniform params: unspecified quantized dimension, when multiple scales
126 // expected-error@+1 {{illegal quantized dimension: -1}}
/llvm-project/mlir/test/CAPI/
H A Dquant.c115 // CHECK: quantized element type: !quant.any<i8<-8:7>:f32> in testAnyQuantizedType()
116 fprintf(stderr, "quantized element type: "); in testAnyQuantizedType()
192 // CHECK: quantized dim: 1 in testUniformPerAxisType()
193 fprintf(stderr, "quantized dim: %" PRId32 "\n", in testUniformPerAxisType()
/llvm-project/mlir/lib/Dialect/Quant/IR/
H A DTypeDetail.h35 // Floating point type that the quantized type approximates.
108 // Floating point type that the quantized type approximates.
176 // Floating point type that the quantized type approximates.
260 // Floating point type that the quantized type approximates.
H A DTypeParser.cpp231 // Optionally parse quantized dimension for per-axis quantization. in parseUniformType()
334 "unknown quantized type " + typeNameSpelling); in parseType()
427 llvm_unreachable("Unhandled quantized type"); in printType()
/llvm-project/mlir/lib/Dialect/Quant/Utils/
H A DFakeQuantSupport.cpp76 // (real value, corresponding quantized value), of which, two such pairs in getNudgedScaleAndZeroPoint()
123 // 0.0s, so the scale is set to 1.0 and the tensor can be quantized to zero in fakeQuantAttrsToType()
/llvm-project/mlir/include/mlir/Dialect/Tosa/IR/
H A DTosaTypesBase.td25 // The base class of a quantized type.
40 // Non-Quantized Signed Integer Types.
60 // Quantized Integer Types.
H A DTosaUtilOps.td36 multiple quantized operations (mul, convolution, rescale, matmul, pooling).
H A DTosaOps.td413 For quantized integer data types, the TABLE operator should be used instead
436 For quantized integer data types, the TABLE operator should be used instead
458 For quantized integer data types, the TABLE operator should be used instead
1930 Rescale quantized values into a new domain. Supported rescalings are:
/llvm-project/mlir/test/Integration/Dialect/SparseTensor/CPU/
H A Dsparse_quantized_matmul.mlir36 // An example of a quantized sparse matmul. With the zero offset for the
/llvm-project/mlir/lib/Dialect/Linalg/Transforms/
H A DTransposeConv2D.cpp48 /// with an analogous example for the quantized case.
/llvm-project/mlir/python/mlir/dialects/linalg/opdsl/ops/
H A Dcore_named_ops.py397 them to the same data type as the accumulator/output. The quantized variant
554 them to the same data type as the accumulator/output. The quantized variant
820 point offsets common to quantized operations.
850 point offsets common to quantized operations.
880 point offsets common to quantized operations.
1013 point offsets common to quantized operations.
1048 point offsets common to quantized operations.
1118 point offsets common to quantized operations.
/llvm-project/mlir/lib/Dialect/Tosa/Transforms/
H A DTosaFolders.cpp216 // TOSA supports quantized types. in matchAndRewrite()
/llvm-project/mlir/lib/Dialect/Tosa/IR/
H A DTosaOps.cpp251 // Either both must be quantized or both unquantized. in verifyConvOp()
259 // Quantized type must have constructed the quantizationattr, and unquantized in verifyConvOp()
263 op.emitOpError("quantizationattr is required for quantized type, and not " in verifyConvOp()
498 assert(inputQType && "Tensor must have quantized datatype!"); in buildPadOpWithQuantInfo()
/llvm-project/mlir/test/Dialect/Tosa/
H A Dinvalid.mlir54 // expected-error@+1 {{'tosa.conv2d' op quantizationattr is required for quantized type, and not allowed for float type}}

12