Media/intel-graphics-compiler
1
0
Fork 0
mirror of https://github.com/intel/intel-graphics-compiler.git synced 2025-11-04 08:21:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

Remaining DX11 & DX12 int64 typed atomics extenions

Browse Source 
Adding remaining DX11 & DX12 int64 typed atomics extenions
min, and, or, add, xor, xchange, compare xchange
This commit is contained in:
Priyanka Patil
2021-12-10 18:29:13 +00:00
committed by igcbot
parent ebd9a8f708
commit dd355c31dc
2 changed files with 186 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

94
inc/IntelExtensions.hlsl
View File

@ -353,7 +353,7 @@ int IntelExt_WavePrefixSum(int value)
return g_IntelExt[opcode].dst0u.x;
}
// uint64 atomics
// uint64 typed atomics
// Interlocked max
uint2 IntelExt_InterlockedMaxUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
@ -366,3 +366,95 @@ uint2 IntelExt_InterlockedMaxUint64(RWTexture2D<uint2> uav, uint2 address, uint2
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked Min
uint2 IntelExt_InterlockedMinUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_MIN;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked and
uint2 IntelExt_InterlockedAndUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_AND;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked or
uint2 IntelExt_InterlockedOrUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_OR;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked add
uint2 IntelExt_InterlockedAddUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_ADD;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked xor
uint2 IntelExt_InterlockedXorUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_XOR;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked exchange
uint2 IntelExt_InterlockedExchangeUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_XCHG;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked compare exchange
uint2 IntelExt_InterlockedCompareExchangeUint64(RWTexture2D<uint2> uav, uint2 address, uint2 cmp_value, uint2 xchg_value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = cmp_value;
g_IntelExt[opcode].src1u.zw = xchg_value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_CMPXCHG;
return g_IntelExt[opcode].dst0u.xy;
}

94
inc/IntelExtensions12.hlsl
View File

@ -72,7 +72,7 @@ void IntelExt_Init()
}
// uint64 atomics
// uint64 typed atomics
// Interlocked max
uint2 IntelExt_InterlockedMaxUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
@ -85,3 +85,95 @@ uint2 IntelExt_InterlockedMaxUint64(RWTexture2D<uint2> uav, uint2 address, uint2
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked min
uint2 IntelExt_InterlockedMinUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_MIN;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked and
uint2 IntelExt_InterlockedAndUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_AND;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked or
uint2 IntelExt_InterlockedOrUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_OR;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked add
uint2 IntelExt_InterlockedAddUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_ADD;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked xor
uint2 IntelExt_InterlockedXorUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_XOR;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked xor
uint2 IntelExt_InterlockedExchangeUint64(RWTexture2D<uint2> uav, uint2 address, uint2 value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_XCHG;
return g_IntelExt[opcode].dst0u.xy;
}
// Interlocked compare exchange
uint2 IntelExt_InterlockedCompareExchangeUint64(RWTexture2D<uint2> uav, uint2 address, uint2 cmp_value, uint2 xchg_value)
{
uint opcode = g_IntelExt.IncrementCounter();
uav[uint2(opcode, opcode)] = uint2(0, 0); //dummy instruction to get the resource handle
g_IntelExt[opcode].opcode = INTEL_EXT_UINT64_ATOMIC;
g_IntelExt[opcode].src0u.xy = address;
g_IntelExt[opcode].src1u.xy = cmp_value;
g_IntelExt[opcode].src1u.zw = xchg_value;
g_IntelExt[opcode].src2u.x = INTEL_EXT_ATOMIC_CMPXCHG;
return g_IntelExt[opcode].dst0u.xy;
}