From 52b78ad9bf99660593c49a8da0bdcec19016ee25 Mon Sep 17 00:00:00 2001
From: jsmall-nvidia <jsmall@nvidia.com>
Date: Fri, 31 Jan 2020 08:55:40 -0500
Subject: 64 bit types doc (#1194)

* * For integer literals add postfix, and use unsigned/signed output appropriately
* Extend GLSL extension handling by type, and for adding 64 bit int extensions
* Added tests for int/uint64 types

* Add explicit Int/UInt64 emit functions to avoid ambiguity.

* Fix uint64_t intrinsics on CUDA/C++.

* WIP 64 bit types documentation.

* Testing int64 intrinsic support.

* Dx12 Dxil sm6.0 does actually support int64_t.
---
 docs/64bit-type-support.md | 78 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
 create mode 100644 docs/64bit-type-support.md

(limited to 'docs/64bit-type-support.md')

diff --git a/docs/64bit-type-support.md b/docs/64bit-type-support.md
new file mode 100644
index 000000000..6b026d3b2
--- /dev/null
+++ b/docs/64bit-type-support.md
@@ -0,0 +1,78 @@
+Slang 64-bit Type Support
+=========================
+
+The Slang language supports 64 bit built in types. Such as
+
+* double
+* uint64_t
+* int64_t
+
+This also applies to vector and matrix versions of these types. 
+
+Unfortunately if a specific target supports the type or the typical HLSL instrinsic functions (such as sin/cos/max/min etc) depends very much on the target. 
+
+Note this initial testing only tested scalar usage, and not vector or matrix intrinsics.
+
+Double support
+==============
+
+Target   | Compiler/Binary  |  Double Type   |   Intrinsics          |  Notes
+---------|------------------|----------------|-----------------------|-----------
+CPU      |                  |      Yes       |          Yes          |  1
+CUDA     | Nvrtx/PTX        |      Yes       |          Yes          |  1
+D3D12    | DXC/DXIL         |      Yes       |          No           |  2 
+Vulkan   | GlSlang/Spir-V   |      Yes       |          No           |  3
+D3D11    | FXC/DXBC         |      No        |          No           |
+D3D12    | FXC/DXBC         |      No        |          No           | 
+
+1) CUDA and CPU support most intrinsics, with the notable exception currently of matrix invert
+2) Requires SM 6.0 and above  https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/hlsl-shader-model-6-0-features-for-direct3d-12
+3) Restriction is described in  https://www.khronos.org/registry/spir-v/specs/1.0/GLSL.std.450.html
+Note that GlSlang does produce spir-v that contains double intrinsic calls, the failure happens when validating the Spir-V 
+
+```
+Validation: error 0:  [ UNASSIGNED-CoreValidation-Shader-InconsistentSpirv ] Object: VK_NULL_HANDLE (Type = 0) | SPIR-V module not valid: GLSL.std.450 Sin: expected Result Type to be a 16 or 32-bit scalar or vector float type
+  %57 = OpExtInst %double %1 Sin %56
+```
+
+D3D12 and VK may have some very limited intrinsic support such as sqrt, rsqrt
+
+uint64_t Support
+=================
+
+Target   | Compiler/Binary  |  uint64_t Type |  Intrinsic support | Notes
+---------|------------------|----------------|--------------------|--------
+CPU      |                  |      Yes       |          Yes       |   
+CUDA     | Nvrtx/PTX        |      Yes       |          Yes       |   
+D3D12    | DXC/DXIL         |      Yes       |          Yes       |   
+Vulkan   | GlSlang/Spir-V   |      Yes       |          Yes       |   
+D3D11    | FXC/DXBC         |      No        |          No        |   1
+D3D12    | FXC/DXBC         |      No        |          No        |   1
+
+1) uint64_t support requires https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/hlsl-shader-model-6-0-features-for-direct3d-12, so DXBC is not a target.
+
+The intrinsics available on uint64_t type are `abs`, `min`, `max`, `clamp` and `countbits`.
+
+int64_t Support
+================
+
+Target   | Compiler/Binary  |  int64_t Type |  Intrinsic support | Notes
+---------|------------------|----------------|--------------------|--------
+CPU      |                  |      Yes       |          Yes       |   
+CUDA     | Nvrtx/PTX        |      Yes       |          Yes       |   
+Vulkan   | GlSlang/Spir-V   |      Yes       |          Yes       |   
+D3D12    | DXC/DXIL         |      Yes       |          Yes       | 1
+D3D11    | FXC/DXBC         |      No        |          No        | 2 
+D3D12    | FXC/DXBC         |      No        |          No        | 2
+
+1) The sm6.0 docs (https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/hlsl-shader-model-6-0-features-for-direct3d-12) describe only supports uint64_t, but the dxc compiler page says int64_t is supported in HLSL 2016 (https://github.com/Microsoft/DirectXShaderCompiler/wiki/Language-Versions). Tests show that this is indeed the case.
+
+2) uint64_t support requires https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/hlsl-shader-model-6-0-features-for-direct3d-12, so DXBC is not a target.
+
+The intrinsics available on uint64_t type are `abs`, `min`, `max` and `clamp`.
+
+GLSL
+====
+
+GLSL/Spir-v based targets do not support 'generated' intrinsics on matrix types. For example 'sin(mat)' will not work on GLSL/Spir-v.
+
-- 
cgit v1.2.3