[Mlir-commits] [mlir] 076f87a - [MLIR][SPIRV] Add support for GLSL F/U/SClamp.
Lei Zhang
llvmlistbot at llvm.org
Sun Dec 13 07:01:53 PST 2020
Author: ergawy
Date: 2020-12-13T09:56:46-05:00
New Revision: 076f87a86741f96c076cea9f9f2af17de55122a3
URL: https://github.com/llvm/llvm-project/commit/076f87a86741f96c076cea9f9f2af17de55122a3
DIFF: https://github.com/llvm/llvm-project/commit/076f87a86741f96c076cea9f9f2af17de55122a3.diff
LOG: [MLIR][SPIRV] Add support for GLSL F/U/SClamp.
Adds support for 3 ternary ops from SPIR-V extended instructions for
GLSL. Namely, adds support for FClamp, UClamp, and SClamp.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D92859
Added:
Modified:
mlir/include/mlir/Dialect/SPIRV/SPIRVBase.td
mlir/include/mlir/Dialect/SPIRV/SPIRVGLSLOps.td
mlir/test/Dialect/SPIRV/Serialization/glsl-ops.mlir
mlir/test/Dialect/SPIRV/glslops.mlir
Removed:
################################################################################
diff --git a/mlir/include/mlir/Dialect/SPIRV/SPIRVBase.td b/mlir/include/mlir/Dialect/SPIRV/SPIRVBase.td
index 54e5efe5f295..0d80ecc8ddda 100644
--- a/mlir/include/mlir/Dialect/SPIRV/SPIRVBase.td
+++ b/mlir/include/mlir/Dialect/SPIRV/SPIRVBase.td
@@ -3075,6 +3075,7 @@ def SPV_Type : AnyTypeOf<[
SPV_AnyCooperativeMatrix, SPV_AnyMatrix
]>;
+def SPV_SignedInt : SignedIntOfWidths<[8, 16, 32, 64]>;
def SPV_SignlessOrUnsignedInt : SignlessOrUnsignedIntOfWidths<[8, 16, 32, 64]>;
class SPV_CoopMatrixOfType<list<Type> allowedTypes> :
diff --git a/mlir/include/mlir/Dialect/SPIRV/SPIRVGLSLOps.td b/mlir/include/mlir/Dialect/SPIRV/SPIRVGLSLOps.td
index 10cafd825116..73745fe4694e 100644
--- a/mlir/include/mlir/Dialect/SPIRV/SPIRVGLSLOps.td
+++ b/mlir/include/mlir/Dialect/SPIRV/SPIRVGLSLOps.td
@@ -77,6 +77,28 @@ class SPV_GLSLBinaryArithmeticOp<string mnemonic, int opcode, Type type,
list<OpTrait> traits = []> :
SPV_GLSLBinaryOp<mnemonic, opcode, type, type, traits>;
+// Base class for GLSL ternary ops.
+class SPV_GLSLTernaryArithmeticOp<string mnemonic, int opcode, Type type,
+ list<OpTrait> traits = []> :
+ SPV_GLSLOp<mnemonic, opcode, !listconcat([NoSideEffect], traits)> {
+
+ let arguments = (ins
+ SPV_ScalarOrVectorOf<type>:$x,
+ SPV_ScalarOrVectorOf<type>:$y,
+ SPV_ScalarOrVectorOf<type>:$z
+ );
+
+ let results = (outs
+ SPV_ScalarOrVectorOf<type>:$result
+ );
+
+ let parser = [{ return impl::parseOneResultSameOperandTypeOp(parser, result); }];
+
+ let printer = [{ return impl::printOneResultOp(getOperation(), p); }];
+
+ let verifier = [{ return success(); }];
+}
+
// -----
def SPV_GLSLFAbsOp : SPV_GLSLUnaryArithmeticOp<"FAbs", 4, SPV_Float> {
@@ -862,4 +884,92 @@ def SPV_GLSLTanhOp : SPV_GLSLUnaryArithmeticOp<"Tanh", 21, SPV_Float16or32> {
}];
}
+// -----
+
+def SPV_GLSLFClampOp : SPV_GLSLTernaryArithmeticOp<"FClamp", 43, SPV_Float> {
+ let summary = "Clamp x between min and max values.";
+
+ let description = [{
+ Result is min(max(x, minVal), maxVal). The resulting value is undefined if
+ minVal > maxVal. The semantics used by min() and max() are those of FMin and
+ FMax.
+
+ The operands must all be a scalar or vector whose component type is
+ floating-point.
+
+ Result Type and the type of all operands must be the same type. Results are
+ computed per component.
+
+ <!-- End of AutoGen section -->
+ ```
+ fclamp-op ::= ssa-id `=` `spv.GLSL.FClamp` ssa-use, ssa-use, ssa-use `:`
+ float-scalar-vector-type
+ ```
+ #### Example:
+
+ ```mlir
+ %2 = spv.GLSL.FClamp %x, %min, %max : f32
+ %3 = spv.GLSL.FClamp %x, %min, %max : vector<3xf16>
+ ```
+ }];
+}
+
+// -----
+
+def SPV_GLSLUClampOp : SPV_GLSLTernaryArithmeticOp<"UClamp", 44, SPV_SignlessOrUnsignedInt> {
+ let summary = "Clamp x between min and max values.";
+
+ let description = [{
+ Result is min(max(x, minVal), maxVal), where x, minVal and maxVal are
+ interpreted as unsigned integers. The resulting value is undefined if
+ minVal > maxVal.
+
+ Result Type and the type of the operands must both be integer scalar or
+ integer vector types. Result Type and operand types must have the same number
+ of components with the same component width. Results are computed per
+ component.
+
+ <!-- End of AutoGen section -->
+ ```
+ uclamp-op ::= ssa-id `=` `spv.GLSL.UClamp` ssa-use, ssa-use, ssa-use `:`
+ unsgined-signless-scalar-vector-type
+ ```
+ #### Example:
+
+ ```mlir
+ %2 = spv.GLSL.UClamp %x, %min, %max : i32
+ %3 = spv.GLSL.UClamp %x, %min, %max : vector<3xui16>
+ ```
+ }];
+}
+
+// -----
+
+def SPV_GLSLSClampOp : SPV_GLSLTernaryArithmeticOp<"SClamp", 45, SPV_SignedInt> {
+ let summary = "Clamp x between min and max values.";
+
+ let description = [{
+ Result is min(max(x, minVal), maxVal), where x, minVal and maxVal are
+ interpreted as signed integers. The resulting value is undefined if
+ minVal > maxVal.
+
+ Result Type and the type of the operands must both be integer scalar or
+ integer vector types. Result Type and operand types must have the same number
+ of components with the same component width. Results are computed per
+ component.
+
+ <!-- End of AutoGen section -->
+ ```
+ uclamp-op ::= ssa-id `=` `spv.GLSL.UClamp` ssa-use, ssa-use, ssa-use `:`
+ sgined-scalar-vector-type
+ ```
+ #### Example:
+
+ ```mlir
+ %2 = spv.GLSL.SClamp %x, %min, %max : si32
+ %3 = spv.GLSL.SClamp %x, %min, %max : vector<3xsi16>
+ ```
+ }];
+}
+
#endif // SPIRV_GLSL_OPS
diff --git a/mlir/test/Dialect/SPIRV/Serialization/glsl-ops.mlir b/mlir/test/Dialect/SPIRV/Serialization/glsl-ops.mlir
index 223b6301207d..d635bde9cbf1 100644
--- a/mlir/test/Dialect/SPIRV/Serialization/glsl-ops.mlir
+++ b/mlir/test/Dialect/SPIRV/Serialization/glsl-ops.mlir
@@ -30,4 +30,22 @@ spv.module Logical GLSL450 requires #spv.vce<v1.0, [Shader], []> {
%12 = spv.GLSL.Round %arg0 : f32
spv.Return
}
+
+ spv.func @fclamp(%arg0 : f32, %arg1 : f32, %arg2 : f32) "None" {
+ // CHECK: spv.GLSL.FClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : f32
+ %13 = spv.GLSL.FClamp %arg0, %arg1, %arg2 : f32
+ spv.Return
+ }
+
+ spv.func @uclamp(%arg0 : ui32, %arg1 : ui32, %arg2 : ui32) "None" {
+ // CHECK: spv.GLSL.UClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : i32
+ %13 = spv.GLSL.UClamp %arg0, %arg1, %arg2 : ui32
+ spv.Return
+ }
+
+ spv.func @sclamp(%arg0 : si32, %arg1 : si32, %arg2 : si32) "None" {
+ // CHECK: spv.GLSL.SClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : si32
+ %13 = spv.GLSL.SClamp %arg0, %arg1, %arg2 : si32
+ spv.Return
+ }
}
diff --git a/mlir/test/Dialect/SPIRV/glslops.mlir b/mlir/test/Dialect/SPIRV/glslops.mlir
index 3e699ed05958..42377c2277a7 100644
--- a/mlir/test/Dialect/SPIRV/glslops.mlir
+++ b/mlir/test/Dialect/SPIRV/glslops.mlir
@@ -269,3 +269,79 @@ func @roundvec(%arg0 : vector<3xf16>) -> () {
%2 = spv.GLSL.Round %arg0 : vector<3xf16>
return
}
+
+// -----
+
+//===----------------------------------------------------------------------===//
+// spv.GLSL.FClamp
+//===----------------------------------------------------------------------===//
+
+func @fclamp(%arg0 : f32, %min : f32, %max : f32) -> () {
+ // CHECK: spv.GLSL.FClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : f32
+ %2 = spv.GLSL.FClamp %arg0, %min, %max : f32
+ return
+}
+
+// -----
+
+func @fclamp(%arg0 : vector<3xf32>, %min : vector<3xf32>, %max : vector<3xf32>) -> () {
+ // CHECK: spv.GLSL.FClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : vector<3xf32>
+ %2 = spv.GLSL.FClamp %arg0, %min, %max : vector<3xf32>
+ return
+}
+
+// -----
+
+//===----------------------------------------------------------------------===//
+// spv.GLSL.UClamp
+//===----------------------------------------------------------------------===//
+
+func @fclamp(%arg0 : ui32, %min : ui32, %max : ui32) -> () {
+ // CHECK: spv.GLSL.UClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : ui32
+ %2 = spv.GLSL.UClamp %arg0, %min, %max : ui32
+ return
+}
+
+// -----
+
+func @fclamp(%arg0 : vector<4xi32>, %min : vector<4xi32>, %max : vector<4xi32>) -> () {
+ // CHECK: spv.GLSL.UClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : vector<4xi32>
+ %2 = spv.GLSL.UClamp %arg0, %min, %max : vector<4xi32>
+ return
+}
+
+// -----
+
+func @fclamp(%arg0 : si32, %min : si32, %max : si32) -> () {
+ // expected-error @+1 {{must be 8/16/32/64-bit signless/unsigned integer or vector}}
+ %2 = spv.GLSL.UClamp %arg0, %min, %max : si32
+ return
+}
+
+// -----
+
+//===----------------------------------------------------------------------===//
+// spv.GLSL.SClamp
+//===----------------------------------------------------------------------===//
+
+func @fclamp(%arg0 : si32, %min : si32, %max : si32) -> () {
+ // CHECK: spv.GLSL.SClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : si32
+ %2 = spv.GLSL.SClamp %arg0, %min, %max : si32
+ return
+}
+
+// -----
+
+func @fclamp(%arg0 : vector<4xsi32>, %min : vector<4xsi32>, %max : vector<4xsi32>) -> () {
+ // CHECK: spv.GLSL.SClamp {{%[^,]*}}, {{%[^,]*}}, {{%[^,]*}} : vector<4xsi32>
+ %2 = spv.GLSL.SClamp %arg0, %min, %max : vector<4xsi32>
+ return
+}
+
+// -----
+
+func @fclamp(%arg0 : i32, %min : i32, %max : i32) -> () {
+ // expected-error @+1 {{must be 8/16/32/64-bit signed integer or vector}}
+ %2 = spv.GLSL.SClamp %arg0, %min, %max : i32
+ return
+}
More information about the Mlir-commits
mailing list