[DirectX] Update "dx.TypedBuffer" docs to include a "signed" bit

[bogner]

To lower these types to dxil we need to know whether ints are signed or not, but the LLVM type loses that. Add a bit to indicate it's so.

[llvmbot]

@llvm/pr-subscribers-backend-directx

Author: Justin Bogner (bogner)

Changes

To lower these types to dxil we need to know whether ints are signed or not, but the LLVM type loses that. Add a bit to indicate it's so.

---

Full diff: https://github.com/llvm/llvm-project/pull/100695.diff

1 Files Affected:

• (modified) llvm/docs/DirectX/DXILResources.rst (+12-9)

diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst
index 5bbe902d9d2d7..8f7b11f7255c0 100644
--- a/llvm/docs/DirectX/DXILResources.rst
+++ b/llvm/docs/DirectX/DXILResources.rst
@@ -96,7 +96,7 @@ Buffers
 
 .. code-block:: llvm
 
-   target("dx.TypedBuffer", ElementType, IsWriteable, IsROV)
+   target("dx.TypedBuffer", ElementType, IsWriteable, IsROV, IsSigned)
    target("dx.RawBuffer", ElementType, IsWriteable, IsROV)
 
 We need two separate buffer types to account for the differences between the
@@ -106,9 +106,10 @@ used for DXIL's RawBuffers and StructuredBuffers. We call the latter
 "RawBuffer" to match the naming of the operations, but it can represent both
 the Raw and Structured variants.
 
-For TypedBuffer, the element type must be an integer or floating point type.
-For RawBuffer the type can be an integer, floating point, or struct type.
-HLSL's ByteAddressBuffer is represented by an `i8` element type.
+For TypedBuffer, the element type must be an scalar integer or floating point
+type, or a vector of at most 4 such types. For RawBuffer the type can be an
+integer, floating point, vecvtor, or struct type. HLSL's ByteAddressBuffer is
+represented as a RawBuffer with an `i8` element type.
 
 These types are generally used by BufferLoad and BufferStore operations, as
 well as atomics.
@@ -128,6 +129,8 @@ There are a few fields to describe variants of all of these types:
        writeable) and UAVs (writeable).
    * - IsROV
      - Whether the UAV is a rasterizer ordered view. Always ``0`` for SRVs.
+   * - IsSigned
+     - Whether the element type is signed ("dx.TypedBuffer" only)
 
 .. _bufferLoad: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#bufferload
 .. _bufferStore: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#bufferstore
@@ -197,23 +200,23 @@ Examples:
 .. code-block:: llvm
 
    ; RWBuffer<float4> Buf : register(u5, space3)
-   %buf = call target("dx.TypedBuffer", float, 1, 0)
+   %buf = call target("dx.TypedBuffer", <4 x float>, 1, 0, 0)
                @llvm.dx.handle.fromBinding.tdx.TypedBuffer_f32_1_0(
                    i32 3, i32 5, i32 1, i32 0, i1 false)
 
-   ; RWBuffer<uint> Buf : register(u7, space2)
-   %buf = call target("dx.TypedBuffer", i32, 1, 0)
+   ; RWBuffer<int> Buf : register(u7, space2)
+   %buf = call target("dx.TypedBuffer", i32, 1, 0, 1)
                @llvm.dx.handle.fromBinding.tdx.TypedBuffer_i32_1_0t(
                    i32 2, i32 7, i32 1, i32 0, i1 false)
 
    ; Buffer<uint4> Buf[24] : register(t3, space5)
-   %buf = call target("dx.TypedBuffer", i32, 0, 0)
+   %buf = call target("dx.TypedBuffer", <4 x i32>, 0, 0, 0)
                @llvm.dx.handle.fromBinding.tdx.TypedBuffer_i32_0_0t(
                    i32 2, i32 7, i32 24, i32 0, i1 false)
 
    ; struct S { float4 a; uint4 b; };
    ; StructuredBuffer<S> Buf : register(t2, space4)
-   %buf = call target("dx.RawBuffer", {<4 x f32>, <4 x i32>}, 0, 0)
+   %buf = call target("dx.RawBuffer", {<4 x float>, <4 x i32>}, 0, 0)
                @llvm.dx.handle.fromBinding.tdx.RawBuffer_sl_v4f32v4i32s_0_0t(
                    i32 4, i32 2, i32 1, i32 0, i1 false)
 

[damyanp]

What would it mean to have an unsigned float? I don't quite understand how this isn't inferable from the element type. Wonder if some more explanation in this doc might be warranted?

[bogner]

Added a couple of sentences to make this clearer

[bob80905]

You say in this doc that IsSigned is an extra parameter, but that param is only present for TypedBuffer. Yet, I see a new parameter here.
Maybe you meant to write somewhere that RWBuffers are represented as TypedBuffers? Should that be made explicit? (Like you specified with ByteAddressBuffers above).
Otherwise, shouldn't we specify that RWBuffers also get an extra parameter, IsSigned?

[bogner]

I thought it was fairly explicit already, but I've added some wording to be clearer about that.