[clang] [CIR][NVPTX] Support WMMA fragment loads (PR #234812)

via cfe-commits cfe-commits at lists.llvm.org
Fri Nov 13 13:46:17 PST 2026


github-actions[bot] wrote:

<!--PREMERGE ADVISOR COMMENT: Windows-->
# :window: Windows x64 Test Results

* 59483 tests passed
* 561 tests skipped
* 1 test failed

## Failed Tests
(click on a test name to see its output)

### Clang
<details>
<summary>Clang.CIR/CodeGenCUDA/builtins-nvptx-wmma.cu</summary>

```
Exit Code: 1

Command Output (stdout):
--
# RUN: at line 1
c:\_work\llvm-project\llvm-project\build\bin\clang.exe -cc1 -internal-isystem C:\_work\llvm-project\llvm-project\build\lib\clang\24\include -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70    -target-feature +ptx60 -x cuda -fcuda-is-device -fclangir    -emit-cir C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu -o - | c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu --check-prefix=CIR --enable-var-scope
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\clang.exe' -cc1 -internal-isystem 'C:\_work\llvm-project\llvm-project\build\lib\clang\24\include' -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70 -target-feature +ptx60 -x cuda -fcuda-is-device -fclangir -emit-cir 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' -o -
# note: command had no output on stdout or stderr
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe' 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' --check-prefix=CIR --enable-var-scope
# note: command had no output on stdout or stderr
# RUN: at line 4
c:\_work\llvm-project\llvm-project\build\bin\clang.exe -cc1 -internal-isystem C:\_work\llvm-project\llvm-project\build\lib\clang\24\include -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70    -target-feature +ptx60 -x cuda -fcuda-is-device -fclangir    -emit-llvm C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu -o - | c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu --check-prefix=LLVM --enable-var-scope
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\clang.exe' -cc1 -internal-isystem 'C:\_work\llvm-project\llvm-project\build\lib\clang\24\include' -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70 -target-feature +ptx60 -x cuda -fcuda-is-device -fclangir -emit-llvm 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' -o -
# note: command had no output on stdout or stderr
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe' 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' --check-prefix=LLVM --enable-var-scope
# note: command had no output on stdout or stderr
# RUN: at line 7
c:\_work\llvm-project\llvm-project\build\bin\clang.exe -cc1 -internal-isystem C:\_work\llvm-project\llvm-project\build\lib\clang\24\include -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70    -target-feature +ptx60 -x cuda -fcuda-is-device    -emit-llvm C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu -o - | c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu --check-prefix=LLVM --enable-var-scope
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\clang.exe' -cc1 -internal-isystem 'C:\_work\llvm-project\llvm-project\build\lib\clang\24\include' -nostdsysteminc -triple nvptx64-nvidia-cuda -target-cpu sm_70 -target-feature +ptx60 -x cuda -fcuda-is-device -emit-llvm 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' -o -
# note: command had no output on stdout or stderr
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe' 'C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu' --check-prefix=LLVM --enable-var-scope
# .---command stderr------------
# | C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu:32:10: error: LLVM: expected string not found in input
# | // LLVM: %[[P1:.*]] = getelementptr{{.*}} i32, ptr %[[DST]], i{{32|64}} 1
# |          ^
# | <stdin>:25:35: note: scanning from here
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:25:35: note: with "DST" equal to "0"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:25:35: note: pattern attempts to capture variables: "P1"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:27:2: note: possible intended match here
# |  %10 = getelementptr i32, ptr %0, i32 2
# |  ^
# | C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu:86:10: error: LLVM: expected string not found in input
# | // LLVM: %[[P1:.*]] = getelementptr{{.*}} i32, ptr %[[DST]], i{{32|64}} 1
# |          ^
# | <stdin>:127:35: note: scanning from here
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:127:35: note: with "DST" equal to "0"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:127:35: note: pattern attempts to capture variables: "P1"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:129:2: note: possible intended match here
# |  %10 = getelementptr i32, ptr %0, i32 2
# |  ^
# | C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu:140:10: error: LLVM: expected string not found in input
# | // LLVM: %[[P1:.*]] = getelementptr{{.*}} i32, ptr %[[DST]], i{{32|64}} 1
# |          ^
# | <stdin>:229:35: note: scanning from here
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:229:35: note: with "DST" equal to "0"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:229:35: note: pattern attempts to capture variables: "P1"
# |  %9 = bitcast <2 x half> %8 to i32
# |                                   ^
# | <stdin>:231:2: note: possible intended match here
# |  %10 = getelementptr i32, ptr %0, i32 2
# |  ^
# | C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu:175:10: error: LLVM: expected string not found in input
# | // LLVM: %[[P1:.*]] = getelementptr{{.*}} float, ptr %[[DST]], i{{32|64}} 1
# |          ^
# | <stdin>:297:84: note: scanning from here
# |  %7 = extractvalue { float, float, float, float, float, float, float, float } %3, 1
# |                                                                                    ^
# | <stdin>:297:84: note: with "DST" equal to "0"
# |  %7 = extractvalue { float, float, float, float, float, float, float, float } %3, 1
# |                                                                                    ^
# | <stdin>:297:84: note: pattern attempts to capture variables: "P1"
# |  %7 = extractvalue { float, float, float, float, float, float, float, float } %3, 1
# |                                                                                    ^
# | <stdin>:299:2: note: possible intended match here
# |  %8 = getelementptr float, ptr %0, i32 2
# |  ^
# | 
# | Input file: <stdin>
# | Check file: C:\_work\llvm-project\llvm-project\clang\test\CIR\CodeGenCUDA\builtins-nvptx-wmma.cu
# | 
# | -dump-input=help explains the following input dump.
# | 
# | Input was:
# | <<<<<<
# |              .
# |              .
# |              .
# |             20:  %5 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 0 
# |             21:  %6 = bitcast <2 x half> %5 to i32 
# |             22:  store i32 %6, ptr %4, align 4 
# |             23:  %7 = getelementptr i32, ptr %0, i32 1 
# |             24:  %8 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 1 
# |             25:  %9 = bitcast <2 x half> %8 to i32 
# | check:32'0                                       {   search range start (exclusive)
# | check:32'1                                           error: no match found in search range
# | check:32'2                                           with "DST" equal to "0"
# | check:32'3                                           pattern attempts to capture variables: "P1"
# |             26:  store i32 %9, ptr %7, align 4 
# |             27:  %10 = getelementptr i32, ptr %0, i32 2 
# | check:32'4       ?                                        possible intended match
# |             28:  %11 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 2 
# |             29:  %12 = bitcast <2 x half> %11 to i32 
# |             30:  store i32 %12, ptr %10, align 4 
# |             31:  %13 = getelementptr i32, ptr %0, i32 3 
# |             32:  %14 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 3 
# |              .
# |              .
# |              .
# |             53:  
# |             54: ; Function Attrs: nocallback nounwind willreturn memory(argmem: read) 
# |             55: declare { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } @llvm.nvvm.wmma.m16n16k16.load.a.row.stride.f16.p0(ptr readonly captures(none), i32) #1 
# |             56:  
# |             57: ; Function Attrs: convergent mustprogress noinline nounwind optnone 
# |             58: define dso_local void @load_a_f16_col(ptr noundef %dst, ptr noundef %src, i32 noundef %stride) #0 { 
# | check:32'5                                            }                                                               search range end (exclusive)
# |             59: entry: 
# |             60:  %dst.addr = alloca ptr, align 8 
# |             61:  %src.addr = alloca ptr, align 8 
# |             62:  %stride.addr = alloca i32, align 4 
# |             63:  store ptr %dst, ptr %dst.addr, align 8 
# |              .
# |              .
# |              .
# |            122:  %5 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 0 
# |            123:  %6 = bitcast <2 x half> %5 to i32 
# |            124:  store i32 %6, ptr %4, align 4 
# |            125:  %7 = getelementptr i32, ptr %0, i32 1 
# |            126:  %8 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 1 
# |            127:  %9 = bitcast <2 x half> %8 to i32 
# | check:86'0                                       {   search range start (exclusive)
# | check:86'1                                           error: no match found in search range
# | check:86'2                                           with "DST" equal to "0"
# | check:86'3                                           pattern attempts to capture variables: "P1"
# |            128:  store i32 %9, ptr %7, align 4 
# |            129:  %10 = getelementptr i32, ptr %0, i32 2 
# | check:86'4       ?                                        possible intended match
# |            130:  %11 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 2 
# |            131:  %12 = bitcast <2 x half> %11 to i32 
# |            132:  store i32 %12, ptr %10, align 4 
# |            133:  %13 = getelementptr i32, ptr %0, i32 3 
# |            134:  %14 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 3 
# |              .
# |              .
# |              .
# |            155:  
# |            156: ; Function Attrs: nocallback nounwind willreturn memory(argmem: read) 
# |            157: declare { <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half> } @llvm.nvvm.wmma.m16n16k16.load.b.row.stride.f16.p0(ptr readonly captures(none), i32) #1 
# |            158:  
# |            159: ; Function Attrs: convergent mustprogress noinline nounwind optnone 
# |            160: define dso_local void @load_b_f16_col(ptr noundef %dst, ptr noundef %src, i32 noundef %stride) #0 { 
# | check:86'5                                            }                                                               search range end (exclusive)
# |            161: entry: 
# |            162:  %dst.addr = alloca ptr, align 8 
# |            163:  %src.addr = alloca ptr, align 8 
# |            164:  %stride.addr = alloca i32, align 4 
# |            165:  store ptr %dst, ptr %dst.addr, align 8 
# |              .
# |              .
# |              .
# |            224:  %5 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 0 
# |            225:  %6 = bitcast <2 x half> %5 to i32 
# |            226:  store i32 %6, ptr %4, align 4 
# |            227:  %7 = getelementptr i32, ptr %0, i32 1 
# |            228:  %8 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 1 
# |            229:  %9 = bitcast <2 x half> %8 to i32 
# | check:140'0                                      {   search range start (exclusive)
# | check:140'1                                          error: no match found in search range
# | check:140'2                                          with "DST" equal to "0"
# | check:140'3                                          pattern attempts to capture variables: "P1"
# |            230:  store i32 %9, ptr %7, align 4 
# |            231:  %10 = getelementptr i32, ptr %0, i32 2 
# | check:140'4      ?                                        possible intended match
# |            232:  %11 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 2 
# |            233:  %12 = bitcast <2 x half> %11 to i32 
# |            234:  store i32 %12, ptr %10, align 4 
# |            235:  %13 = getelementptr i32, ptr %0, i32 3 
# |            236:  %14 = extractvalue { <2 x half>, <2 x half>, <2 x half>, <2 x half> } %3, 3 
# |              .
# |              .
# |              .
# |            241:  
# |            242: ; Function Attrs: nocallback nounwind willreturn memory(argmem: read) 
# |            243: declare { <2 x half>, <2 x half>, <2 x half>, <2 x half> } @llvm.nvvm.wmma.m16n16k16.load.c.row.stride.f16.p0(ptr readonly captures(none), i32) #1 
# |            244:  
# |            245: ; Function Attrs: convergent mustprogress noinline nounwind optnone 
# |            246: define dso_local void @load_c_f16_col(ptr noundef %dst, ptr noundef %src, i32 noundef %stride) #0 { 
# | check:140'5                                           }                                                               search range end (exclusive)
# |            247: entry: 
# |            248:  %dst.addr = all
# | ...
# `---data was truncated (10240/13069) (change limit with -D output_limit=N)
# error: command failed with exit status: 1

--

```
</details>

If these failures are unrelated to your changes (for example tests are broken or flaky at HEAD), please open an issue at https://github.com/llvm/llvm-project/issues and add the `infrastructure` label.

https://github.com/llvm/llvm-project/pull/234812


More information about the cfe-commits mailing list