test: delete the extra cast in cstyle load [run_process_replay] [no_assert] #5310
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Changes |
ignaciosica
added a commit
to ignaciosica/tinygrad
that referenced
this pull request
Jul 9, 2024
commit 51882c8efcd033fb13b3c50f1a80d4be4672a99a
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Tue Jul 9 14:53:41 2024 -0300
Squashed commit of the following:
commit 1678199
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Tue Jul 9 20:44:44 2024 +0300
add update_copy to hcq spec (tinygrad#5348)
* add update_copy to hcq spec
* fix amd
commit 9504db1
Author: chenyu <chenyu@fastmail.com>
Date: Tue Jul 9 12:28:52 2024 -0400
remove the realize in _rebuild_tensor_v2 (tinygrad#5347)
no longer needed
commit 1f5de80
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Tue Jul 9 17:20:17 2024 +0300
multi reduce Tensor.var passing verify_lazyop (tinygrad#5346)
* what about this
* reset late gate
commit 3d45219
Author: kormann <49917710+DKormann@users.noreply.github.com>
Date: Tue Jul 9 15:38:39 2024 +0200
[bug fix] nested commutative pattern _match [run_process_replay] [no_assert] (tinygrad#5340)
* deep pat test
* lint
* min diff
* min lines
* nothing
* is res extra
* cleanup2
* add res back
* reduce lines
* type anno
---------
Co-authored-by: qazal <qazal.software@gmail.com>
commit e815c57
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Tue Jul 9 15:56:06 2024 +0300
use hcq_profile in nv/amd program (tinygrad#5344)
commit bee96a1
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Tue Jul 9 15:24:56 2024 +0300
fuzz uop schedules (tinygrad#5345)
* basic blocks + cleanups
* fixups
* elif is better for future me
* fuzz_schedule_max_paths
* fix linter
commit d5a68ae
Author: Ian Paul <46036106+ianpaul10@users.noreply.github.com>
Date: Tue Jul 9 10:48:42 2024 +0000
Simple abstractions3.py fix (tinygrad#5343)
* abstractions3.py fix
* Add abstractions3.py to CI tests
commit a2a9bfd
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Tue Jul 9 10:39:39 2024 +0300
nv correct error messages with ptx (tinygrad#5341)
* nv correct error messages with ptx
* return compile error
commit c13da83
Author: George Hotz <72895+geohot@users.noreply.github.com>
Date: Mon Jul 8 21:23:19 2024 -0700
tests from lowerer branch (tinygrad#5339)
* tests from lowerer branch
* Update test_image_dtype.py
* Update test_image_dtype.py
* Update test_image_dtype.py
commit 4ceab5d
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 22:25:03 2024 -0400
fix PTX match rule for gated LOAD (tinygrad#5338)
* test padto sum with bool tensor and bool acc dtype
make sure bool tensor acc with gate is handled correctly
* broken in PTX
* fix ptx
commit a80f2df
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 21:33:05 2024 -0400
fix some PTX tests (tinygrad#5337)
fix broken PTX tests in test_linearizer and test_uops. there are tests that were skipped and broken because it runs only with CUDA=1 and we run PTX with NV=1 now
commit 9150a6b
Author: wozeparrot <wozeparrot@gmail.com>
Date: Tue Jul 9 00:45:40 2024 +0000
tensor metadata (tinygrad#5271)
commit 7f642aa
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 19:19:20 2024 -0400
minor PTX matcher cleanup [run_process_replay] (tinygrad#5336)
* minor PTX matcher cleanup [run_process_replay]
uop.cast syntatic sugar and some newline/space cleanup
* comment
commit 0f09402
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 18:15:04 2024 -0400
fix Tensor.all and Tensor.any for PTX (tinygrad#5335)
supported boolean acc and boolean phi. and rewrite boolean max to uint8 max
commit 053c706
Author: Roelof van Dijk <3604013+roelofvandijk@users.noreply.github.com>
Date: Mon Jul 8 20:47:34 2024 +0200
refactor: expr_view on View (tinygrad#5315)
commit 2349d83
Author: kormann <49917710+DKormann@users.noreply.github.com>
Date: Mon Jul 8 20:46:15 2024 +0200
Fix scope order in graph toposort [run_process_replay] (tinygrad#5330)
* fix
* test
* nothing
commit 631bc97
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 14:00:28 2024 -0400
raise line count limit to 8500 (tinygrad#5331)
commit bb77469
Author: Timmy <96938750+0xtimmy@users.noreply.github.com>
Date: Mon Jul 8 10:28:55 2024 -0700
multireduce scheduler tests (tinygrad#5141)
Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com>
commit bb2222e
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Mon Jul 8 19:01:27 2024 +0300
nv default for ampere & ada (tinygrad#5329)
commit 51d6f37
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Mon Jul 8 18:25:05 2024 +0300
nv get classes based on device (tinygrad#5325)
* nv get classes
* support in mockgpu
* choose sm based on gpu
* fix
* fix
* fix arch
commit 7d049fc
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 8 10:51:56 2024 -0400
move getting 0 and min value of a dtype to dtype.py (tinygrad#5328)
cleanup getting base case for reduce ops
[run_process_replay]
commit b0c5c58
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Mon Jul 8 17:24:33 2024 +0300
nv rm_control to rmctrl type (tinygrad#5327)
* nv rm_control to rmctrl type
* fix
commit 73bddc4
Author: Elias Wahl <82230675+Eliulm@users.noreply.github.com>
Date: Mon Jul 8 15:07:44 2024 +0200
Fix fake dataloader (tinygrad#5326)
commit 43ec8d7ed8dfd420af72875730232a5b2faf3d6b
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Tue Jul 9 14:44:59 2024 -0300
remove comments
commit d1874a53b51859f5d74a7919773c99a535e89c44
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Mon Jul 8 14:13:29 2024 -0300
remove unnecessary code
commit ef5497f141c2f5b48d89707cc77b1b6b02587105
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Mon Jul 8 13:27:50 2024 -0300
remove simple_relu and env
commit 61be81b3fa1d57ea612a97c72e53430f369bf59f
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Mon Jul 8 13:26:37 2024 -0300
working with accs
commit 8803069c0459a276d373d4c0c0feeb0b3b057761
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Mon Jul 8 11:02:05 2024 -0300
make amx an OptOps
commit 49119c2b62d89eaa0a2825714b44b549f7145edb
Merge: 97faa1c3 6856f91
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Sun Jul 7 15:55:03 2024 -0300
merge master
commit 6856f91
Author: chenyu <chenyu@fastmail.com>
Date: Sun Jul 7 14:36:00 2024 -0400
Tensor.any and Tensor.all (tinygrad#5320)
does not work in ptx yet due to how boolean tensor is handled
commit 2029cb7
Author: chenyu <chenyu@fastmail.com>
Date: Sun Jul 7 13:04:22 2024 -0400
support passing None to Tensor.clip (tinygrad#5319)
passing None for no upper bound or no lower bound
commit 296a1a3
Author: chenyu <chenyu@fastmail.com>
Date: Sun Jul 7 12:10:39 2024 -0400
update Tensor.round doc and example (tinygrad#5318)
document rounding half to even and update examples to show
commit c1e330f
Author: chenyu <chenyu@fastmail.com>
Date: Sun Jul 7 11:52:58 2024 -0400
Tensor.int and Tensor.bool (tinygrad#5317)
commit 778d1cd
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Sun Jul 7 17:34:49 2024 +0300
nv allocate local memory dynamically (tinygrad#5277)
* nv allocate local memory dynamically
* fix
* linter
* linter 2
* linter
* fixes
commit ae10e93
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sun Jul 7 10:49:08 2024 +0300
UOps.VECTORIZE cleanups [run_process_replay] (tinygrad#5314)
* still render_cast
* one extra line ok
* these are all just vectorize
* save space
* behavior change can go in a different diff
commit 77b2ce9
Author: greg-niemeyer <152219575+greg-niemeyer@users.noreply.github.com>
Date: Sat Jul 6 23:59:57 2024 -0700
Add UOps.VECTORIZE [run_process_replay] (tinygrad#5289)
* Add UOps.VECTORIZE to core
* Update vectorized cast tests
* Addresses code review comments
- Removes VECTORIZE from LLVMRenderer
- Add line breaks to unduly long lines
- Add noop CAST rule back
- Update asserts and add render_vectorize in
CSytleLanguage renderer
* Add missing const folding rule for VECTORIZE
Also adds corresponding test
* Fixes test_const_vectorize_fold and add assert
- Use sane types with VECTORIZE in test_const_vectorize_fold
- Add assert that sanity checks the types for VECTORIZE
* Rename test_cast_vectorized_fold
Renames test_cast_vectorized_fold to test_noop_vectorize_fold
because the test targets a very specific rule and there are
other tests for VECTORIZE.
* Revert unrelated changes
---------
Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com>
Co-authored-by: qazal <qazal.software@gmail.com>
commit 2a7282c
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sun Jul 7 09:12:49 2024 +0300
test: delete the extra cast in cstyle load [run_process_replay] [no_assert] (tinygrad#5310)
* test: delete the extra cast in cstyle load [run_process_replay] [no_assert]
* assert buf_uop
* ImageDType
* ptx is actually a 64bit address
commit cededd8
Author: chenyu <chenyu@fastmail.com>
Date: Sat Jul 6 21:55:59 2024 -0400
minor multi cleanup (tinygrad#5311)
add type, move around and some newlines
commit 8a99514
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sun Jul 7 00:06:30 2024 +0300
generalize the uops toposort spec to ptx (tinygrad#5309)
* generalize spec to ptx
* redundant assert
* extra print
commit ca0ef17
Author: chenyu <chenyu@fastmail.com>
Date: Sat Jul 6 12:47:27 2024 -0400
use precise::sin in metal (tinygrad#5307)
commit 5c2ca7b
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sat Jul 6 19:28:47 2024 +0300
remove UOps.SPECIAL rendering from llvm (tinygrad#5306)
commit 356e5d2
Author: chenyu <chenyu@fastmail.com>
Date: Sat Jul 6 11:54:12 2024 -0400
touchup multi dtype in elementwise (tinygrad#5305)
only need to check real once, also added type annotation
commit 7ddda9f
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sat Jul 6 14:13:58 2024 +0300
hotfix: cache seen graphs in fusion (tinygrad#5302)
commit 11dfb19
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sat Jul 6 12:39:31 2024 +0300
track seen graphs in recursive group (tinygrad#5301)
* track seen
* maybe never add realized
* ahh it needs to track sts
* delete extra check
* cache typings
* minor cleanup
commit d813617
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sat Jul 6 12:04:03 2024 +0300
prescheduling refactor (tinygrad#5300)
* p1
* refactor tuple
commit c1e166c
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Sat Jul 6 11:36:40 2024 +0300
fix dtype mismatch for bool ops in multi (tinygrad#5299)
commit fc03fc0
Author: chenyu <chenyu@fastmail.com>
Date: Fri Jul 5 14:52:09 2024 -0400
enable sin on METAL in test_dtype_alu (tinygrad#5298)
commit b369e75
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Fri Jul 5 21:14:38 2024 +0300
refactor schedule creation (tinygrad#5297)
commit 5292d37
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Fri Jul 5 19:43:50 2024 +0300
LoadOps.VIEW in the scheduler spec (tinygrad#5296)
* refactor to allow_buffer_view
* tests
* fix multi
commit 1ab7a4c
Author: hikettei <88639579+hikettei@users.noreply.github.com>
Date: Sat Jul 6 01:16:44 2024 +0900
Handling Multiple UnaryOps.BITCAST in Function for Proper Kernel Fusion [run_process_replay] (tinygrad#5172)
* [Patch] added an option not to ignore view replacing when doing bitcast
* added the testcase
* [Add] reproduced bitcast cannot be fused into a single kernel in the unittest
---------
Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com>
commit 43c3f73
Author: chenyu <chenyu@fastmail.com>
Date: Fri Jul 5 11:01:20 2024 -0400
handcode_bert_opt.py (tinygrad#5295)
similar to handcode_resnet50_opt.py, one file to check bert kernels without dataset.
commit d7835a7
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Fri Jul 5 16:53:40 2024 +0300
hotfix: fix metal with vars (tinygrad#5294)
* hotfix: fix metal with vars
* one more place
commit 8a548b0
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Fri Jul 5 16:13:05 2024 +0300
metal support offset (tinygrad#5293)
commit 1cefbb3
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Fri Jul 5 13:00:01 2024 +0300
uop graph tests + type_verify cleanup (tinygrad#5292)
* test_cast_alu_fold
* test_double_cast_fold + these should assert
commit 341c4a2
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Fri Jul 5 11:29:35 2024 +0300
hotfix: use dtype.scalar() for rendering cast [run_process_replay] [no_assert] (tinygrad#5290)
commit 87d27c4
Author: chenyu <chenyu@fastmail.com>
Date: Thu Jul 4 14:25:24 2024 -0400
minor _broadcast cleanup (tinygrad#5286)
`any(x==0 for x in y)` is `0 in y`.
also `get_args(ConstType)` instead of hard coded `float, int, bool`
commit 8c03816
Author: SnakeOnex <romakcz@gmail.com>
Date: Thu Jul 4 17:15:07 2024 +0200
fix README example (tinygrad#5284)
* fixed README example
* README test
* changed py -> python markdown code flags in REAME
commit 2778b60
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Thu Jul 4 18:06:04 2024 +0300
new memory scheduler (tinygrad#5278)
* new memory schedule algo
* works
* fix
* fix
* linter
* tiny fixes
* do not optimize copy buffers
* mpre comments
* tiny cleanups
commit 84b3e3b
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Thu Jul 4 13:29:21 2024 +0300
hcq exec no embedded signal (tinygrad#5142)
commit 0c3a35e
Author: Tobias Fischer <tobiasfischer17@gmail.com>
Date: Wed Jul 3 22:47:10 2024 -0400
Stable Diffusion v2 Inference (tinygrad#5283)
* model implementation
* clip fix, more qol options
commit e5ba385
Author: chenyu <chenyu@fastmail.com>
Date: Wed Jul 3 19:42:56 2024 -0400
remove first contiguous in multi from_sharded (tinygrad#5121)
second contiguous guarantees lbs are contiguous going into MultiLazyBuffer, don't need the first contiguous
commit f1ff65e
Author: chenyu <chenyu@fastmail.com>
Date: Wed Jul 3 17:52:50 2024 -0400
remove "no-nans-fp-math"="true" for LLVM (tinygrad#5282)
fixed isnan for llvm (still have issue with < nan)
commit 97faa1c307ee9322a88de83cf6c2d2ae8918db56
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Wed Jul 3 16:01:58 2024 -0300
minor changes
commit 3929a9d
Author: chenyu <chenyu@fastmail.com>
Date: Wed Jul 3 14:59:05 2024 -0400
fix UOp.cmp_tuple for ALU (tinygrad#5280)
* fix UOp.cmp_tuple for ALU
for ALU, use self.arg instead of self.op to compare
* skip that?
commit a9d6a6c
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Wed Jul 3 20:15:42 2024 +0300
verify_lazyop with multi reduce (tinygrad#5276)
* outsource the assert to the implicit movement op check
* tests
commit 16e3b8b
Author: George Hotz <72895+geohot@users.noreply.github.com>
Date: Wed Jul 3 09:40:00 2024 -0700
uops work from lowerer [run_process_replay] (tinygrad#5279)
commit 622b7bd
Author: chenyu <chenyu@fastmail.com>
Date: Wed Jul 3 12:28:53 2024 -0400
simpler TinyJit inside TinyJit detection (tinygrad#5219)
* simpler TinyJit inside TinyJit detection
suggested in tinygrad@73395b9#commitcomment-143660402
* cannot repro...
* clear the way out
* finally clear
commit 04ef0fd
Author: gip <gip@users.noreply.github.com>
Date: Wed Jul 3 09:07:09 2024 -0700
fix: message when applegpu tools missiong (tinygrad#5236)
commit d3e244d
Author: reddyn12 <72528507+reddyn12@users.noreply.github.com>
Date: Wed Jul 3 12:06:01 2024 -0400
prev speed improvements (tinygrad#5252)
Co-authored-by: reddyn <nikidsniper@gmail.com>
commit 8c90b4d4b0f4d0ba97d3e9ee12d3253523e69bca
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Wed Jul 3 11:25:30 2024 -0300
vec is done before store
commit 21d41f0
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Wed Jul 3 11:34:10 2024 +0300
nv follows HCQCompatAllocRes protocol (tinygrad#5275)
* nv follows HCQCompatAllocRes protocol
* fix amd
commit d3e4e21
Author: Vyacheslav Pachkov <slava.pach@gmail.com>
Date: Wed Jul 3 10:25:44 2024 +0300
add return type for HCQCompatAllocator _alloc (tinygrad#5267)
Co-authored-by: nimlgen <138685161+nimlgen@users.noreply.github.com>
commit 191463a
Author: chenyu <chenyu@fastmail.com>
Date: Tue Jul 2 23:29:54 2024 -0400
add timing to SDXL (tinygrad#5273)
commit b2c3a28
Author: chenyu <chenyu@fastmail.com>
Date: Tue Jul 2 21:39:01 2024 -0400
nn.RMSNorm (tinygrad#5272)
the norm itself has no significant value to add to Tensor method, but we would want Tensor.normalize
commit 9a2a82a
Author: chenyu <chenyu@fastmail.com>
Date: Tue Jul 2 21:37:52 2024 -0400
test stable diffusion unet in ci (tinygrad#5268)
unet is parameterized now so can test a smaller one is ci
commit ce52b10
Author: chenyu <chenyu@fastmail.com>
Date: Tue Jul 2 20:01:11 2024 -0400
add a flag DISABLE_LOOP_COLLAPSE (tinygrad#5270)
workaround if user encountered UNMUL error
commit e53b164
Author: George Hotz <72895+geohot@users.noreply.github.com>
Date: Tue Jul 2 15:03:54 2024 -0700
small changes from lowerer (tinygrad#5266)
commit 7be776f
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Tue Jul 2 23:35:39 2024 +0300
add _alloc_signal/_free_signal to hcq (tinygrad#5264)
* add _alloc_signal/_free_signal api
* oops, revert this
* linter
commit 9a25ee0
Author: Tobias Fischer <tobiasfischer17@gmail.com>
Date: Tue Jul 2 12:40:27 2024 -0400
pixed unet call params (tinygrad#5262)
commit b4cb32bac18b7ca2426e243c40fc90fe90b014ad
Author: p4sscode <p4ssenger.developer@gmail.com>
Date: Tue Jul 2 10:41:20 2024 -0300
squash merge amx_support for third revision
commit 59bc837
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Tue Jul 2 15:13:10 2024 +0300
refactor gated load rendering [run_process_replay] (tinygrad#5259)
* refactor gated load rendering [run_process_replay]
* hotfix: extra line
* remove llvm diff
commit e050603
Author: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Tue Jul 2 13:57:46 2024 +0300
nv close fds after mapping (tinygrad#5246)
commit d3cfb6c
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Tue Jul 2 13:48:47 2024 +0300
refactor UOps.LOAD barrier [run_process_replay] (tinygrad#5258)
commit a1044e6
Author: qazal <77887910+Qazalin@users.noreply.github.com>
Date: Tue Jul 2 09:21:09 2024 +0300
iterate over scoped uops once [run_process_replay] (tinygrad#5255)
commit dfbee4f
Author: wozeparrot <wozeparrot@gmail.com>
Date: Tue Jul 2 02:33:58 2024 +0000
feat: add blobfile to testing (tinygrad#5254)
commit 8c9c1cf
Author: Tobias Fischer <tobiasfischer17@gmail.com>
Date: Mon Jul 1 22:33:01 2024 -0400
Pulled CLIP and UNet into Seperate Files (tinygrad#5253)
* pulled clip and unet into seperate files
* reference cleanup, lru cache fix
* better pool indexing
commit 5808c37
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 1 15:00:47 2024 -0400
hotfix disable flaky llama3 beam benchmark on green (tinygrad#5249)
commit b9122ec
Author: chenyu <chenyu@fastmail.com>
Date: Mon Jul 1 14:43:47 2024 -0400
revert stable diffusion validation with threefry (tinygrad#5248)
* Revert "use threefry in stable diffusion benchmark (tinygrad#4988)"
This reverts commit 44dfa37.
* sdxl and validation fix
* relax threshold
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
CI proves in cstyle backends a float4 load is already the correct dtype.
also, LOAD's buf_dtype always matches its output_dtype.scalar() (except for PTX where src[0].dtype is rewritten as u64).