llvm-beanz commented 11 months ago

AC:

The intrinsic is implemented
Tests validate that it is lowered to LLVM-IR
Tests validate that control forms are not hoisted out of control flow ** Maps to the correct construct
[ ] Implement GroupMemoryBarrierWithGroupSync clang builtin,
[ ] Link GroupMemoryBarrierWithGroupSync clang builtin with hlsl_intrinsics.h
[ ] Add sema checks for GroupMemoryBarrierWithGroupSync to CheckHLSLBuiltinFunctionCall in SemaChecking.cpp
[ ] Add codegen for GroupMemoryBarrierWithGroupSync to EmitHLSLBuiltinExpr in CGBuiltin.cpp
[ ] Add codegen tests to clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
[ ] Add sema tests to clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrierWithGroupSync-errors.hlsl
[ ] Create the int_dx_GroupMemoryBarrierWithGroupSync intrinsic in IntrinsicsDirectX.td
[ ] Create the DXILOpMapping of int_dx_GroupMemoryBarrierWithGroupSync to 80 in DXIL.td
[ ] Create the GroupMemoryBarrierWithGroupSync.ll and GroupMemoryBarrierWithGroupSync_errors.ll tests in llvm/test/CodeGen/DirectX/
[ ] Create the int_spv_GroupMemoryBarrierWithGroupSync intrinsic in IntrinsicsSPIRV.td
[ ] In SPIRVInstructionSelector.cpp create the GroupMemoryBarrierWithGroupSync lowering and map it to int_spv_GroupMemoryBarrierWithGroupSync in SPIRVInstructionSelector::selectIntrinsic.
[ ] Create SPIR-V backend test case in llvm/test/CodeGen/SPIRV/hlsl-intrinsics/GroupMemoryBarrierWithGroupSync.ll

DirectX

DXIL Opcode	DXIL OpName	Shader Model	Shader Stages
80	Barrier	6.0	()

SPIR-V

OpControlBarrier:

Description:

Wait for all active invocations within the specified Scope to reach the current point of execution.

All active invocations within Execution scope reach this point of execution before any invocation proceeds beyond it.

When Execution is Workgroup or larger, behavior is undefined unless all invocations within Execution execute the same dynamic instance of this instruction.

If Semantics is not None, this instruction also serves as an OpMemoryBarrier instruction, and also performs and adheres to the description and semantics of an OpMemoryBarrier instruction with the same Memory and Semantics operands. This allows atomically specifying both a control barrier and a memory barrier (that is, without needing two instructions). If Semantics is None, Memory is ignored.

Before version 1.3, it is only valid to use this instruction with TessellationControl, GLCompute, or Kernel execution models. There is no such restriction starting with version 1.3.

If used with the TessellationControl execution model, it also implicitly synchronizes the Output Storage Class: Writes to Output variables performed by any invocation executed prior to a OpControlBarrier are visible to any other invocation proceeding beyond that OpControlBarrier.

Word Count	Opcode	Results	Operands
4	224	Scope <id> Execution	Scope <id> Memory	Memory Semantics <id> Semantics

Test Case(s)

Example 1

//dxc GroupMemoryBarrierWithGroupSync_test.hlsl -T lib_6_8  -enable-16bit-types -spirv -fspv-target-env=universal1.5 -fcgl -O0

export void fn() {
    return GroupMemoryBarrierWithGroupSync();
}

HLSL:

Blocks execution of all threads in a group until all group shared accesses have been completed and all threads in the group have reached this call.

Syntax

void GroupMemoryBarrierWithGroupSync(void);

Parameters

This function has no parameters.

Return value

This function does not return a value.

Remarks

The behavior of calls to this function that are within diverging branches of a thread group are undefined.

Minimum Shader Model

This function is supported in the following shader models.

Shader Model	Supported
Shader Model 5 and higher shader models	yes

This function is supported in the following types of shaders:

Vertex	Hull	Domain	Geometry	Pixel	Compute
					x

llvm / llvm-project

Implement the `GroupMemoryBarrierWithGroupSync` HLSL Function #70103