Add subscript operator support in cuda #6830

mkeshavaNV · 2025-04-16T12:10:07Z

This CL adds support for the subscript operator for Read Only
textures in cuda. Also adds a test for this.

Now that we support .Load for read only textures in cuda, an existing test also needs to be modified.

mkeshavaNV · 2025-04-25T10:56:34Z

/format

slangbot · 2025-04-25T10:57:26Z

🌈 Formatted, please merge the changes from this PR

…h-6781 Format code for PR shader-slang#6830

Copilot

Pull Request Overview

This PR adds support for the subscript operator for read-only textures in CUDA by introducing inline PTX intrinsic functions to perform texture fetch/load operations across 1D, 2D, 3D, and array textures. Key changes include:

Implementation of tex1Dfetch_int, tex2Dfetch_int, tex3Dfetch_int functions for various dimensional textures.
Addition of tex1DArrayfetch_int and tex2DArrayfetch_int to support array textures.
Updates to the test suite to validate the new functionality.

Files not reviewed (4)

source/slang/hlsl.meta.slang: Language not supported
tests/compute/texture-subscript-cuda.slang: Language not supported
tests/compute/texture-subscript-cuda.slang.expected.txt: Language not supported
tests/language-feature/capability/capability-invalid-fragment-in-compute.slang: Language not supported

skallweitNV · 2025-04-25T11:59:48Z

source/slang/hlsl.meta.slang

+                    switch(Shape.flavor)
+                    {
+                    case $(SLANG_TEXTURE_1D):
+                        __intrinsic_asm "<PTX does not support 1D texture read>";


isn't this just going to lead to an error in the downstream compiler (nvrtc)?
should that be a static_assert instead?

makes sense. updated.

hm, seems like the bootstrap has some issues with the assert. I will try to fix that.

ok seems like that was because of missing default. I've added that back, and the static_assert. Please check.

skallweitNV · 2025-04-25T12:01:14Z

source/slang/hlsl.meta.slang

+                        __intrinsic_asm "tex2Dfetch_int<$T0>($0, ($1).x, ($1).y)";
+                    case $(SLANG_TEXTURE_3D):
+                        __intrinsic_asm "tex3Dfetch_int<$T0>($0, ($1).x, ($1).y, ($1).z)";
+					case $(SLANG_TEXTURE_CUBE):


are there no PTX intrinsics for cube map reads?

nope, cubemap requires float coordinates. Confirmed with the ptx team as well. I even saw error:
line 84; error : Floating-point coordinates required

…h-6781 Format code for PR shader-slang#6830

This CL adds support for the subscript operator for Read Only textures in cuda. Also adds a test for this. Fixes shader-slang#6781

szihs · 2025-04-28T13:00:27Z

prelude/slang-cuda-prelude.h

+    asm("tex.3d.v4.f32.s32 {%0, %1, %2, %3}, [%4, {%5, %6, %7, %8}];"
+        : "=f"(result), "=f"(dummy), "=f"(dummy), "=f"(dummy)
+        : "l"(texObj), "r"(x), "r"(y), "r"(z), "r"(z));
+    // Note: The repeated z is a WAR for a bug in PTX. That is a dummy parameter.


can you add more details. Does PTX only support vec4 fetch?

Can we write a test which will start failing when this bug is fixed downstream?

oh that is not a bug. Fixed the note after I confirmed with the ptx team. It is how the 3D textures are addressed.

From the docs:
https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#texture-instructions-tex

Operand c is a scalar or singleton tuple for 1d textures; is a two-element vector for 2d textures; and is a four-element vector for 3d textures,

expipiplus1 · 2025-04-28T14:40:08Z

prelude/slang-cuda-prelude.h

+    asm("tex.3d.v4.f32.s32 {%0, %1, %2, %3}, [%4, {%5, %6, %7, %8}];"
+        : "=f"(result), "=f"(dummy), "=f"(dummy), "=f"(dummy)
+        : "l"(texObj), "r"(x), "r"(y), "r"(z), "r"(z));
+    // Note: The repeated z is a WAR for a bug in PTX. That is a dummy parameter.


Can we write a test which will start failing when this bug is fixed downstream?

expipiplus1 · 2025-04-28T14:41:26Z

source/slang/hlsl.meta.slang

+
+                    switch(Shape.flavor)
+                    {
+                    case $(SLANG_TEXTURE_2D):


Do we need to handle 3d here as well, inline with the static assert (I think that an if/else here would be clearer)

I don't think we support 3D array textures in slang? I could not find Texture3DArray. I have updated the static_assert to be clearer, and also made this an if. Please let me know if this is OK.

mkeshavaNV · 2025-04-30T07:29:46Z

/format

slangbot · 2025-04-30T07:30:31Z

🌈 Formatted, please merge the changes from this PR

…h-6781 Format code for PR shader-slang#6830

* cuda: Add support for subscript operator This CL adds support for the subscript operator for Read Only textures in cuda. Also adds a test for this. Fixes shader-slang#6781 * format code * fix review comments * format code --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com> Co-authored-by: Ellie Hermaszewska <ellieh@nvidia.com>

mkeshavaNV changed the title ~~Bugfix/gh 6781~~ WiP: Add Load support for ReadOnly textures in CUDA Apr 16, 2025

mkeshavaNV force-pushed the bugfix/gh-6781 branch from 72b90b2 to 0bd47d4 Compare April 16, 2025 12:17

mkeshavaNV force-pushed the bugfix/gh-6781 branch 2 times, most recently from 795637a to 1fbd035 Compare April 24, 2025 12:16

mkeshavaNV changed the title ~~WiP: Add Load support for ReadOnly textures in CUDA~~ Add Load support for ReadOnly textures in CUDA Apr 24, 2025

mkeshavaNV added the pr: non-breaking PRs without breaking changes label Apr 24, 2025

mkeshavaNV force-pushed the bugfix/gh-6781 branch from 1fbd035 to a9b6b9d Compare April 25, 2025 10:43

slangbot mentioned this pull request Apr 25, 2025

Format code for PR #6830 mkeshavaNV/slang#17

Merged

mkeshavaNV added a commit to mkeshavaNV/slang that referenced this pull request Apr 25, 2025

Merge pull request #17 from slangbot/format-6830-bugfix/shader-slangg…

d3459d8

…h-6781 Format code for PR shader-slang#6830

mkeshavaNV requested a review from expipiplus1 April 25, 2025 11:11

mkeshavaNV marked this pull request as ready for review April 25, 2025 11:11

mkeshavaNV requested a review from a team as a code owner April 25, 2025 11:11

mkeshavaNV requested a review from Copilot April 25, 2025 11:39

Copilot AI reviewed Apr 25, 2025

View reviewed changes

mkeshavaNV changed the title ~~Add Load support for ReadOnly textures in CUDA~~ Add subscript operator support in cuda Apr 25, 2025

skallweitNV reviewed Apr 25, 2025

View reviewed changes

mkeshavaNV added a commit to mkeshavaNV/slang that referenced this pull request Apr 25, 2025

Merge pull request #17 from slangbot/format-6830-bugfix/shader-slangg…

c83cd9a

…h-6781 Format code for PR shader-slang#6830

mkeshavaNV force-pushed the bugfix/gh-6781 branch from d3459d8 to c83cd9a Compare April 25, 2025 12:27

mkeshavaNV and others added 2 commits April 25, 2025 18:05

cuda: Add support for subscript operator

508e6be

This CL adds support for the subscript operator for Read Only textures in cuda. Also adds a test for this. Fixes shader-slang#6781

format code

9dd729e

mkeshavaNV force-pushed the bugfix/gh-6781 branch from c83cd9a to 9dd729e Compare April 25, 2025 12:48

mkeshavaNV requested a review from skallweitNV April 25, 2025 14:51

szihs reviewed Apr 28, 2025

View reviewed changes

expipiplus1 requested changes Apr 28, 2025

View reviewed changes

fix review comments

db27b3c

format code

c8365ad

slangbot mentioned this pull request Apr 30, 2025

Format code for PR #6830 mkeshavaNV/slang#19

Merged

Merge pull request #19 from slangbot/format-6830-bugfix/shader-slangg…

3502327

…h-6781 Format code for PR shader-slang#6830

mkeshavaNV mentioned this pull request Apr 30, 2025

CUDA: Do not silently emit invalid code #6782

Closed

expipiplus1 approved these changes Apr 30, 2025

View reviewed changes

Merge branch 'master' into bugfix/shader-slanggh-6781

ace9e87

mkeshavaNV enabled auto-merge (squash) April 30, 2025 09:29

mkeshavaNV merged commit b0e1505 into shader-slang:master Apr 30, 2025
16 checks passed

skallweitNV mentioned this pull request May 5, 2025

[CUDA] Texture reads emit invalid code #7002

Closed

This was referenced May 8, 2025

cuda: Add more formats for texture read/write #7012

Merged

cuda: Emit error on subscript operator usage for textures #6813

Closed

mkeshavaNV mentioned this pull request May 16, 2025

Issue with Texture2D Subscript Operator Generating No Code and No Error in CUDA Backend #6419

Closed

Add subscript operator support in cuda #6830

Add subscript operator support in cuda #6830

Uh oh!

Conversation

mkeshavaNV commented Apr 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mkeshavaNV commented Apr 25, 2025

Uh oh!

slangbot commented Apr 25, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mkeshavaNV commented Apr 30, 2025

Uh oh!

slangbot commented Apr 30, 2025

Uh oh!

Uh oh!

Uh oh!

mkeshavaNV commented Apr 16, 2025 •

edited

Loading