subject:"\[clang\] \[llvm\] \[mlir\] \[AMDGPU\] add clamp immediate operand to WMMA iu8 intrinsic \(PR #171069\)"

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-27 Thread LLVM Continuous Integration via cfe-commits


llvm-ci wrote:

LLVM Buildbot has detected a new failure on builder `ppc64le-mlir-rhel-clang` 
running on `ppc64le-mlir-rhel-test` while building `clang,llvm,mlir` at step 3 
"clean-build-dir".

Full details are available at: 
https://lab.llvm.org/buildbot/#/builders/129/builds/35561


Here is the relevant piece of the build log for the reference

```
Step 3 (clean-build-dir) failure: Delete failed. (failure) (timed out)
Step 4 (cmake-configure) failure: cmake (failure) (timed out)
command timed out: 1200 seconds without output running [b'cmake', 
b'-DLLVM_TARGETS_TO_BUILD=PowerPC', b'-DLLVM_INSTALL_UTILS=ON', 
b'-DCMAKE_CXX_STANDARD=17', b'-DLLVM_ENABLE_PROJECTS=mlir', 
b'-DLLVM_LIT_ARGS=-vj 256', b'-DCMAKE_C_COMPILER_LAUNCHER=ccache', 
b'-DCMAKE_CXX_COMPILER_LAUNCHER=ccache', b'-DCMAKE_BUILD_TYPE=Release', 
b'-DLLVM_ENABLE_ASSERTIONS=ON', b'-GNinja', b'../llvm-project/llvm'], 
attempting to kill
process killed by signal 9
program finished with exit code -1
elapsedTime=1200.339650

```



https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-27 Thread Shilei Tian via cfe-commits


https://github.com/shiltian closed 
https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-27 Thread Muhammad Abdul via cfe-commits


0xzre wrote:

Hello @shiltian I think no more review incoming and PR been opened for too 
long. Kindly help merge this PR 🙏 , sorry if you're still in holiday/sth

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-22 Thread Muhammad Abdul via cfe-commits


0xzre wrote:

Updated PR description

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-22 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre edited https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-22 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre edited https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-22 Thread Krzysztof Drewniak via cfe-commits


https://github.com/krzysz00 approved this pull request.

MLIR side is good

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Shilei Tian via cfe-commits


https://github.com/shiltian approved this pull request.

LLVM part LGTM. You might want someone else to look at the MLIR part.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From b7a391c60ef7965ee047f6c54308f53dd5e404cf Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/13] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def   |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl|  4 ++--
 ...tins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td   |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td|  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll|  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll |  8 
 .../llvm.amdgcn.wmma.imod.gfx1250.w32.ll   |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td   | 11 ++-
 mlir/test/Target/LLVMIR/rocdl.mlir | 18 +++---
 11 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 88b306462a92c..1851b6a354671 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -857,7 +857,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Muhammad Abdul via cfe-commits



@@ -1665,6 +1665,13 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)

0xzre wrote:

No problem, it happens 👍. I was not being confident enough to push the 
discussion here lol, my bad. Reverting.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Shilei Tian via cfe-commits



@@ -857,7 +857,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")

shiltian wrote:

Hmm, I thought one point, your PR was adding some new builtins instead of 
updating existing ones. That doesn't seem like the case.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Shilei Tian via cfe-commits



@@ -1665,6 +1665,13 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)

shiltian wrote:

When I made that comment, I thought your PR was adding new builtins. That is 
why I said “new builtins”. If that is not the case, your previous handling was 
fine. Could you revert `11b1f67a372a20aafc91f94ce8623db2100a1624`? Sorry for 
the confusion.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From b7a391c60ef7965ee047f6c54308f53dd5e404cf Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/15] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def   |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl|  4 ++--
 ...tins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td   |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td|  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll|  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll |  8 
 .../llvm.amdgcn.wmma.imod.gfx1250.w32.ll   |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td   | 11 ++-
 mlir/test/Target/LLVMIR/rocdl.mlir | 18 +++---
 11 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 88b306462a92c..1851b6a354671 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -857,7 +857,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From b7a391c60ef7965ee047f6c54308f53dd5e404cf Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/14] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def   |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl|  4 ++--
 ...tins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td   |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td|  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll|  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll |  8 
 .../llvm.amdgcn.wmma.imod.gfx1250.w32.ll   |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td   | 11 ++-
 mlir/test/Target/LLVMIR/rocdl.mlir | 18 +++---
 11 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 88b306462a92c..1851b6a354671 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -857,7 +857,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-21 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From b7a391c60ef7965ee047f6c54308f53dd5e404cf Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/13] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def   |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl|  4 ++--
 ...tins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td   |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td|  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll|  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll |  8 
 .../llvm.amdgcn.wmma.imod.gfx1250.w32.ll   |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td   | 11 ++-
 mlir/test/Target/LLVMIR/rocdl.mlir | 18 +++---
 11 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 88b306462a92c..1851b6a354671 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -857,7 +857,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-19 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From c9d94c0002cfd640455d59f720aee2c831f69063 Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/13] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a consta

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-19 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From c9d94c0002cfd640455d59f720aee2c831f69063 Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/13] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a consta

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-19 Thread Shilei Tian via cfe-commits



@@ -1665,6 +1665,13 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)

shiltian wrote:

Builtins are builtins, intrinsics are intrinsics. I don't see what's the issue 
here.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-19 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/12] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a consta

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-19 Thread Muhammad Abdul via cfe-commits



@@ -1665,6 +1665,13 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)

0xzre wrote:

Forcing an 8th arg immediately would break downstream code just as @arsenm 
mentioned here 
https://github.com/llvm/llvm-project/pull/171069#pullrequestreview-3552446843

Maybe what we need is to make clamp required later in a transition plan 🤔 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-18 Thread Shilei Tian via cfe-commits



@@ -1665,6 +1665,13 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)

shiltian wrote:

I'm not sure if that's a good idea to make it an optional argument. They are 
new builtins so I think we just want them to be a regular `i1` argument. In 
that case, you don't need any special handling here and other places.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-18 Thread Shilei Tian via cfe-commits



@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIb.", "nc", "gfx1250-insts,wavefrontsize32")

shiltian wrote:

Oh interesting. I didn't even realize that.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-18 Thread Matt Arsenault via cfe-commits



@@ -0,0 +1,21 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+
+; Verify that the legacy WMMA IU8 intrinsic without the clamp operand is
+; upgraded by appending clamp=false.
+
+define <8 x i32> @wmma_legacy(<8 x i32> %a, <8 x i32> %b, <8 x i32> %c) {
+; CHECK-LABEL: @wmma_legacy(
+; CHECK-NEXT: call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 
false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c, i1 false, i1 false, 
i1 false) #1, !annotation !0
+; CHECK-NEXT: ret <8 x i32>
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(
+  i1 false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c,
+  i1 false, i1 false) #1, !annotation !0
+  ret <8 x i32> %res
+}
+
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(
+  i1, <8 x i32>, i1, <8 x i32>, <8 x i32>, i1, i1)
+
+attributes #1 = { cold }
+
+!0 = !{!"wmma-upgrade"}

arsenm wrote:

```suggestion
attributes #0 = { "preserve-me" }

!0 = !{!"preserve-me"}
```

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-18 Thread Matt Arsenault via cfe-commits



@@ -0,0 +1,21 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+
+; Verify that the legacy WMMA IU8 intrinsic without the clamp operand is
+; upgraded by appending clamp=false.
+
+define <8 x i32> @wmma_legacy(<8 x i32> %a, <8 x i32> %b, <8 x i32> %c) {
+; CHECK-LABEL: @wmma_legacy(
+; CHECK-NEXT: call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 
false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c, i1 false, i1 false, 
i1 false) #1, !annotation !0
+; CHECK-NEXT: ret <8 x i32>
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(
+  i1 false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c,
+  i1 false, i1 false) #1, !annotation !0

arsenm wrote:

```suggestion
  i1 false, i1 false) #0, !annotation !0
```

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-17 Thread Muhammad Abdul via cfe-commits



@@ -108,10 +108,16 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, false, 
mod);

0xzre wrote:

fixed it, please review the implementation. thanks!

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-17 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/11] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a consta

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Juan Manuel Martinez Caamaño via cfe-commits



@@ -108,10 +108,16 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, false, 
mod);

jmmartinez wrote:

Since the llvm-intrinsic expects an immediate, yes I'd rather raise an error in 
the frontend if the clamp argument is not an immediate.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits



@@ -108,10 +108,16 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, false, 
mod);

0xzre wrote:

My implementation is following just like the other WMMA/SWMMAC reuse flags, 
which are allowed to be non-literal. Though I can add `BuiltinConstantArg` on 
`SemaAMDGPU` if needed 👍 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Matt Arsenault via cfe-commits



@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIb.", "nc", "gfx1250-insts,wavefrontsize32")

arsenm wrote:

No, those correspond directly to an opcode 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Juan Manuel Martinez Caamaño via cfe-commits



@@ -108,10 +108,16 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, false, 
mod);

jmmartinez wrote:

Is expected that the test pass on this case when the clamp argument is not a 
literal ?

Should we check for it on `SemaAMDGPU` ? 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 01/10] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a consta

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread via cfe-commits


github-actions[bot] wrote:


# :window: Windows x64 Test Results

* 128696 tests passed
* 2821 tests skipped
* 1 test failed

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

### Clang

Clang.CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl

```
Exit Code: 2

Command Output (stdout):
--
# RUN: at line 3
c:\_work\llvm-project\llvm-project\build\bin\clang.exe -cc1 -internal-isystem 
C:\_work\llvm-project\llvm-project\build\lib\clang\22\include -nostdsysteminc 
-triple amdgcn-unknown-unknown -target-cpu gfx1250 -target-feature 
+wavefrontsize32 -emit-llvm -o - 
C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl
 | c:\_work\llvm-project\llvm-project\build\bin\filecheck.exe 
C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl
 --check-prefix=CHECK-GFX1250
# executed command: 'c:\_work\llvm-project\llvm-project\build\bin\clang.exe' 
-cc1 -internal-isystem 
'C:\_work\llvm-project\llvm-project\build\lib\clang\22\include' -nostdsysteminc 
-triple amdgcn-unknown-unknown -target-cpu gfx1250 -target-feature 
+wavefrontsize32 -emit-llvm -o - 
'C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl'
# .---command stderr
# | Assertion failed: (i >= FTy->getNumParams() || FTy->getParamType(i) == 
Args[i]->getType()) && "Calling a function with a bad signature!", file 
C:\_work\llvm-project\llvm-project\llvm\lib\IR\Instructions.cpp, line 766
# | PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ 
and include the crash backtrace, preprocessed source, and associated run script.
# | Stack dump:
# | 0.  Program arguments: 
c:\\_work\\llvm-project\\llvm-project\\build\\bin\\clang.exe -cc1 
-internal-isystem 
C:\\_work\\llvm-project\\llvm-project\\build\\lib\\clang\\22\\include 
-nostdsysteminc -triple amdgcn-unknown-unknown -target-cpu gfx1250 
-target-feature +wavefrontsize32 -emit-llvm -o - 
C:\\_work\\llvm-project\\llvm-project\\clang\\test\\CodeGenOpenCL\\builtins-amdgcn-gfx1250-wmma-w32.cl
# | 1.  
C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl:166:1:
 current parser token 'void'
# | 2.  
C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl:155:6:
 LLVM IR generation of declaration 'test_amdgcn_wmma_i32_16x16x64_iu8'
# | 3.  
C:\_work\llvm-project\llvm-project\clang\test\CodeGenOpenCL\builtins-amdgcn-gfx1250-wmma-w32.cl:155:6:
 Generating code for declaration 'test_amdgcn_wmma_i32_16x16x64_iu8'
# | Exception Code: 0xC01D
# |  #0 0x7ff752866296 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x7e6296)
# |  #1 0x7ff86387bb04 (C:\Windows\System32\ucrtbase.dll+0x7bb04)
# |  #2 0x7ff86387cad1 (C:\Windows\System32\ucrtbase.dll+0x7cad1)
# |  #3 0x7ff86387e4a1 (C:\Windows\System32\ucrtbase.dll+0x7e4a1)
# |  #4 0x7ff86387e6e1 (C:\Windows\System32\ucrtbase.dll+0x7e6e1)
# |  #5 0x7ff752f32818 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0xeb2818)
# |  #6 0x7ff75259794c 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x51794c)
# |  #7 0x7ff7525973f6 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x5173f6)
# |  #8 0x7ff758a33b55 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x69b3b55)
# |  #9 0x7ff7581aaef1 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x612aef1)
# | #10 0x7ff7581c6de3 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x6146de3)
# | #11 0x7ff756a6385e 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x49e385e)
# | #12 0x7ff75801b036 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x5f9b036)
# | #13 0x7ff75802119c 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x5fa119c)
# | #14 0x7ff7580128a1 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x5f928a1)
# | #15 0x7ff758003503 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x5f83503)
# | #16 0x7ff756a34a0c 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x49b4a0c)
# | #17 0x7ff756a34911 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x49b4911)
# | #18 0x7ff756ce52c0 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x4c652c0)
# | #19 0x7ff756cf3d85 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x4c73d85)
# | #20 0x7ff7569fe9d5 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x497e9d5)
# | #21 0x7ff75476bc34 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x26ebc34)
# | #22 0x7ff754761eff 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x26e1eff)
# | #23 0x7ff75476813a 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x26e813a)
# | #24 0x7ff754760794 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x26e0794)
# | #25 0x7ff757935a8b 
(c:\_work\llvm-project\llvm-project\build\bin\clang.exe+0x58b5a8b)
# | #26 0x

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/9] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/8] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits



@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIb.", "nc", "gfx1250-insts,wavefrontsize32")

0xzre wrote:

Btw are these what you're referring to?
> BUILTIN(__builtin_amdgcn_rsq_clamp, "dd", "nc")
BUILTIN(__builtin_amdgcn_rsq_clampf, "ff", "nc")

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits



@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIb.", "nc", "gfx1250-insts,wavefrontsize32")

0xzre wrote:

Thanks for the suggestion! AFAIK introducing it would duplicate surface area 
and require extra maintenance, without adding new capability. I think we should 
use the current intrinsic unless there's strong use case 🤔

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Shilei Tian via cfe-commits



@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIb.", "nc", "gfx1250-insts,wavefrontsize32")

shiltian wrote:

Alternatively, you can add a builtin 
`__builtin_amdgcn_wmma_i32_16x16x64_iu8_clamp` as well as a corresponding 
intrinsic `llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32.clamp` dedicated for 
the case where clamp is enabled. I saw some other code like that.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits



@@ -1665,6 +1665,24 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)
+Args.push_back(Builder.getFalse());
+
+  auto ToBool = [&](Value *V) {
+if (V->getType()->isIntegerTy(1))
+  return V;
+return Builder.CreateIntCast(V, Builder.getInt1Ty(), false);

0xzre wrote:

Understood, was worried if something upstream left a wider integer.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-16 Thread Muhammad Abdul via cfe-commits



@@ -4504,6 +4511,21 @@ static Value *upgradeARMIntrinsicCall(StringRef Name, 
CallBase *CI, Function *F,
 //
 static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
  Function *F, IRBuilder<> &Builder) {
+  if (Name.starts_with("wmma.i32.16x16x64.iu8")) {
+// Legacy WMMA IU8 intrinsic lacked the optional clamp operand. Append
+// clamp=false for compatibility.
+if (CI->arg_size() != 7)
+  return nullptr;
+
+SmallVector Args(CI->args().begin(), CI->args().end());
+Args.push_back(Builder.getFalse());
+
+Function *NewDecl = Intrinsic::getOrInsertDeclaration(
+  F->getParent(), Intrinsic::amdgcn_wmma_i32_16x16x64_iu8,
+  {CI->getArgOperand(4)->getType(), CI->getArgOperand(1)->getType()});
+return Builder.CreateCall(NewDecl, Args);

0xzre wrote:

the operand list can’t be expanded in place based on `CallBase`. Now will copy 
everything else so behavior is preserved.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -4504,6 +4511,21 @@ static Value *upgradeARMIntrinsicCall(StringRef Name, 
CallBase *CI, Function *F,
 //
 static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
  Function *F, IRBuilder<> &Builder) {
+  if (Name.starts_with("wmma.i32.16x16x64.iu8")) {

arsenm wrote:

Can you make this check based on intrinsic ID? The intrinsic name didn't change 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -4504,6 +4511,21 @@ static Value *upgradeARMIntrinsicCall(StringRef Name, 
CallBase *CI, Function *F,
 //
 static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
  Function *F, IRBuilder<> &Builder) {
+  if (Name.starts_with("wmma.i32.16x16x64.iu8")) {
+// Legacy WMMA IU8 intrinsic lacked the optional clamp operand. Append
+// clamp=false for compatibility.
+if (CI->arg_size() != 7)

arsenm wrote:

Can move the arg size check above the name check 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+
+; Verify that the legacy WMMA IU8 intrinsic without the clamp operand is
+; upgraded by appending clamp=false.
+
+define <8 x i32> @wmma_legacy(<8 x i32> %a, <8 x i32> %b, <8 x i32> %c) {
+; CHECK-LABEL: @wmma_legacy(
+; CHECK: call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 
false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c, i1 false, i1 false, 
i1 false)
+; CHECK: ret <8 x i32>

arsenm wrote:

CHECK-NEXTs 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -1665,6 +1665,24 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)
+Args.push_back(Builder.getFalse());
+
+  auto ToBool = [&](Value *V) {
+if (V->getType()->isIntegerTy(1))
+  return V;
+return Builder.CreateIntCast(V, Builder.getInt1Ty(), false);
+  };
+
+  // Ensure predicate-like operands are i1 to match intrinsic signature.
+  Args[0] = ToBool(Args[0]);
+  Args[2] = ToBool(Args[2]);
+  Args[5] = ToBool(Args[5]);
+  Args[6] = ToBool(Args[6]);
+  Args[7] = ToBool(Args[7]);

arsenm wrote:

Shouldn't need to do anything to these arguments 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -1219,6 +1220,12 @@ static bool upgradeIntrinsicFunction1(Function *F, 
Function *&NewFn,
 break; // No other 'amdgcn.atomic.*'
   }
 
+  if (Name.starts_with("wmma.i32.16x16x64.iu8") && F->arg_size() == 7) {

arsenm wrote:

Swap checks (but also since this wasn't a removed intrinsic, I would expect you 
can still use intrinsic ID checks) 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -4504,6 +4511,21 @@ static Value *upgradeARMIntrinsicCall(StringRef Name, 
CallBase *CI, Function *F,
 //
 static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
  Function *F, IRBuilder<> &Builder) {
+  if (Name.starts_with("wmma.i32.16x16x64.iu8")) {
+// Legacy WMMA IU8 intrinsic lacked the optional clamp operand. Append
+// clamp=false for compatibility.
+if (CI->arg_size() != 7)
+  return nullptr;
+
+SmallVector Args(CI->args().begin(), CI->args().end());
+Args.push_back(Builder.getFalse());
+
+Function *NewDecl = Intrinsic::getOrInsertDeclaration(
+  F->getParent(), Intrinsic::amdgcn_wmma_i32_16x16x64_iu8,
+  {CI->getArgOperand(4)->getType(), CI->getArgOperand(1)->getType()});
+return Builder.CreateCall(NewDecl, Args);

arsenm wrote:

Do you really need to create a new call, and not mutate the one in place? As it 
is this is losing callsite attributes and metadata 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -1665,6 +1665,24 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 if (AppendFalseForOpselArg)
   Args.push_back(Builder.getFalse());
 
+if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
+  if (Args.size() == 7)
+Args.push_back(Builder.getFalse());
+
+  auto ToBool = [&](Value *V) {
+if (V->getType()->isIntegerTy(1))
+  return V;
+return Builder.CreateIntCast(V, Builder.getInt1Ty(), false);

arsenm wrote:

This shouldn't require casting here, this should have been an implicit cast in 
the AST to start with 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-14 Thread Matt Arsenault via cfe-commits



@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+
+; Verify that the legacy WMMA IU8 intrinsic without the clamp operand is
+; upgraded by appending clamp=false.
+
+define <8 x i32> @wmma_legacy(<8 x i32> %a, <8 x i32> %b, <8 x i32> %c) {
+; CHECK-LABEL: @wmma_legacy(
+; CHECK: call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 
false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c, i1 false, i1 false, 
i1 false)
+; CHECK: ret <8 x i32>
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(
+  i1 false, <8 x i32> %a, i1 false, <8 x i32> %b, <8 x i32> %c,
+  i1 false, i1 false)

arsenm wrote:

Test with metadata attachment, and a callsite attribute 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-13 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/7] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-13 Thread via cfe-commits


github-actions[bot] wrote:




:warning: C/C++ code formatter, clang-format found issues in your code. 
:warning:



You can test this locally with the following command:


``bash
git-clang-format --diff origin/main HEAD --extensions cpp,cl -- 
clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp clang/lib/Sema/SemaAMDGPU.cpp 
clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
llvm/lib/IR/AutoUpgrade.cpp --diff_from_common_commit
``

:warning:
The reproduction instructions above might return results for more than one PR
in a stack if you are using a stacked PR workflow. You can limit the results by
changing `origin/main` to the base branch/commit you want to compare against.
:warning:





View the diff from clang-format here.


``diff
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index acb9df20b..7fbfdb722 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -32,8 +32,8 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/IntrinsicsAArch64.h"
-#include "llvm/IR/IntrinsicsARM.h"
 #include "llvm/IR/IntrinsicsAMDGPU.h"
+#include "llvm/IR/IntrinsicsARM.h"
 #include "llvm/IR/IntrinsicsNVPTX.h"
 #include "llvm/IR/IntrinsicsRISCV.h"
 #include "llvm/IR/IntrinsicsWebAssembly.h"
@@ -4612,8 +4612,8 @@ static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, 
CallBase *CI,
 Args.push_back(Builder.getFalse());
 
 Function *NewDecl = Intrinsic::getOrInsertDeclaration(
-  F->getParent(), Intrinsic::amdgcn_wmma_i32_16x16x64_iu8,
-  {CI->getArgOperand(4)->getType(), CI->getArgOperand(1)->getType()});
+F->getParent(), Intrinsic::amdgcn_wmma_i32_16x16x64_iu8,
+{CI->getArgOperand(4)->getType(), CI->getArgOperand(1)->getType()});
 return Builder.CreateCall(NewDecl, Args);
   }
 

``




https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-13 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/6] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-09 Thread Krzysztof Drewniak via cfe-commits


krzysz00 wrote:

> Not yet, where's the preferred doc location? 🤔

Well, if it's listed in the public bits of the ISA manual, it's probably fine, 
on further thought

Otherwise ... that's an annoyingly hard question. My intuitions say AMDGPUUsage

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-09 Thread Muhammad Abdul via cfe-commits


0xzre wrote:

> Documentation somewhere of what `clamp` does?

Not yet, where's the preferred doc location? 🤔 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-09 Thread Muhammad Abdul via cfe-commits

0xzre wrote:

> This can't just break existing uses. Needs to implement bitcode autoupgrade, 
> and some compatibility on the builtin signature.
> 
> Also, can we just infer clamping as an optimization fold based on the use 
> context?

Good catch. Will add a bitcode autupgrade for the old 4 op WMMA IU8 intrinsic 
to append clamp=0, plus a lit bitcode test. I'll keep source compatibility by 
keeping the existing builtin spelling mapped to clamp=0 and adding the 5 op 
form. Clamp inference can be a follow-up combine I think... this patch stays on 
correctness/compat. Would that be okay @arsenm ?

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-08 Thread Krzysztof Drewniak via cfe-commits


https://github.com/krzysz00 commented:

Documentation somewhere of what `clamp` does?

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-08 Thread Matt Arsenault via cfe-commits


https://github.com/arsenm commented:

This can't just break existing uses. Needs to implement bitcode autoupgrade, 
and some compatibility on the builtin signature. 

Also, can we just infer clamping as an optimization fold based on the use 
context? 

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-08 Thread Muhammad Abdul via cfe-commits


0xzre wrote:

hi @shiltian , requesting again help to review. thanks!

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-08 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/4] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread Shilei Tian via cfe-commits



@@ -483,14 +483,19 @@ v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], 
v[16:23] neg_lo:[0,1,0]
 // GFX1250: v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] 
neg_lo:[0,1,0] ; encoding: [0x10,0x00,0x72,0xcc,0x00,0x11,0x42,0x5c]
 // WAVESIZE-ERR: :[[@LINE-3]]:1: error: instruction requires wavesize=32
 
-v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] matrix_a_reuse
+v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] clamp

shiltian wrote:

add the corresponding dasm test as well

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre updated 
https://github.com/llvm/llvm-project/pull/171069

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH 1/2] [AMDGPU] add clamp immediate operand to WMMA iu8
 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, mod, false); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant 
integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, mod); // 
expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a constant

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread Shilei Tian via cfe-commits


https://github.com/shiltian edited 
https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread Shilei Tian via cfe-commits


https://github.com/shiltian commented:

The LLVM changes look good to me.

https://github.com/llvm/llvm-project/pull/171069
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread via cfe-commits


github-actions[bot] wrote:


# :penguin: Linux x64 Test Results

* 166853 tests passed
* 2914 tests skipped
* 4 tests failed

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

### LLVM

LLVM.CodeGen/AMDGPU/wmma-coececution-valu-hazards.mir

```
Exit Code: 2

Command Output (stdout):
--
# RUN: at line 2
/home/gha/actions-runner/_work/llvm-project/llvm-project/build/bin/llc 
-mtriple=amdgcn -mcpu=gfx1250 -run-pass post-RA-hazard-rec 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/test/CodeGen/AMDGPU/wmma-coececution-valu-hazards.mir
 -o - | 
/home/gha/actions-runner/_work/llvm-project/llvm-project/build/bin/FileCheck 
-check-prefix=GFX1250 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/test/CodeGen/AMDGPU/wmma-coececution-valu-hazards.mir
# executed command: 
/home/gha/actions-runner/_work/llvm-project/llvm-project/build/bin/llc 
-mtriple=amdgcn -mcpu=gfx1250 -run-pass post-RA-hazard-rec 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/test/CodeGen/AMDGPU/wmma-coececution-valu-hazards.mir
 -o -
# .---command stderr
# | 
# | # Machine code for function test_wmma_I32_16x16x64_IU8_D0_overlaps_Use1: 
IsSSA, NoPHIs, NoVRegs, TiedOpsRewritten
# | 
# | bb.0:
# |   early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = 
V_WMMA_I32_16X16X64_IU8_w32_twoaddr 8, killed 
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed 
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23(tied-def 0), 0, 0, 0, 
implicit $exec
# |   $vgpr32 = V_ADD_F32_e32 $vgpr16, $vgpr33, implicit $mode, implicit $exec
# | 
# | # End machine code for function test_wmma_I32_16x16x64_IU8_D0_overlaps_Use1.
# | 
# | *** Bad machine code: Expected immediate, but got non-immediate ***
# | - function:test_wmma_I32_16x16x64_IU8_D0_overlaps_Use1
# | - basic block: %bb.0  (0x2e6bb238)
# | - instruction: early-clobber 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = 
V_WMMA_I32_16X16X64_IU8_w32_twoaddr 8, killed 
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed 
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23(tied-def 0), 0, 0, 0, 
implicit $exec
# | 
# | *** Bad machine code: Explicit operand marked as implicit ***
# | - function:test_wmma_I32_16x16x64_IU8_D0_overlaps_Use1
# | - basic block: %bb.0  (0x2e6bb238)
# | - instruction: early-clobber 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = 
V_WMMA_I32_16X16X64_IU8_w32_twoaddr 8, killed 
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed 
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23(tied-def 0), 0, 0, 0, 
implicit $exec
# | - operand 9:   implicit $exec
# | 
# | *** Bad machine code: Expected a non-register operand. ***
# | - function:test_wmma_I32_16x16x64_IU8_D0_overlaps_Use1
# | - basic block: %bb.0  (0x2e6bb238)
# | - instruction: early-clobber 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = 
V_WMMA_I32_16X16X64_IU8_w32_twoaddr 8, killed 
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed 
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed 
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23(tied-def 0), 0, 0, 0, 
implicit $exec
# | - operand 9:   implicit $exec
# | LLVM ERROR: Found 3 machine code errors.
# | PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ 
and include the crash backtrace and instructions to reproduce the bug.
# | Stack dump:
# | 0.  Program arguments: 
/home/gha/actions-runner/_work/llvm-project/llvm-project/build/bin/llc 
-mtriple=amdgcn -mcpu=gfx1250 -run-pass post-RA-hazard-rec 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/test/CodeGen/AMDGPU/wmma-coececution-valu-hazards.mir
 -o -
# |  #0 0x07f40428 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/lib/Support/Unix/Signals.inc:834:13
# |  #1 0x07f3db35 llvm::sys::RunSignalHandlers() 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/lib/Support/Signals.cpp:105:18
# |  #2 0x07f411f1 SignalHandler(int, siginfo_t*, void*) 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/lib/Support/Unix/Signals.inc:426:38
# |  #3 0x7cf334a3a330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
# |  #4 0x7cf334a93b2c pthread_kill 
(/lib/x86_64-linux-gnu/libc.so.6+0x9eb2c)
# |  #5 0x7cf334a3a27e raise (/lib/x86_64-linux-gnu/libc.so.6+0x4527e)
# |  #6 0x7cf334a1d8ff abort (/lib/x86_64-linux-gnu/libc.so.6+0x288ff)
# |  #7 0x07ea4965 llvm::report_fatal_error(llvm::Twine const&, bool) 
/home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/lib/Support/ErrorHandling.cpp:137:5
# |  #8 0x0706e34b 
(/home/gha/actions-runner/_work/llvm-project/llvm-project/build/bin/llc+0x706e34b)
#

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread via cfe-commits


llvmbot wrote:




@llvm/pr-subscribers-llvm-ir

Author: Muhammad Abdul  (0xzre)


Changes

Fixes #166989 

- Adds a clamp immediate operand to the AMDGPU WMMA iu8 intrinsic and threads 
it through LLVM IR, MIR lowering, Clang builtins/tests, and MLIR ROCDL dialect 
so all layers agree on the new operand
- Updates AMDGPUWmmaIntrinsicModsAB so the clamp attribute is emitted, teaches 
VOP3P encoding to accept the immediate, and adjusts Clang codegen/builtin 
headers plus MLIR op definitions and tests to match

---

Patch is 21.73 KiB, truncated to 20.00 KiB below, full version: 
https://github.com/llvm/llvm-project/pull/171069.diff


11 Files Affected:

- (modified) clang/include/clang/Basic/BuiltinsAMDGPU.def (+1-1) 
- (modified) clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
(+2-2) 
- (modified) 
clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl (+5-4) 
- (modified) llvm/include/llvm/IR/IntrinsicsAMDGPU.td (+2-1) 
- (modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+3-1) 
- (modified) llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll (+3-3) 
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll (+2-2) 
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll 
(+4-4) 
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll 
(+3-3) 
- (modified) mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td (+3-2) 
- (modified) mlir/test/Target/LLVMIR/rocdl.mlir (+10-6) 


``diff
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32

[clang] [llvm] [mlir] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic (PR #171069)

2025-12-07 Thread Muhammad Abdul via cfe-commits


https://github.com/0xzre created 
https://github.com/llvm/llvm-project/pull/171069

Fixes #166989 

- Adds a clamp immediate operand to the AMDGPU WMMA iu8 intrinsic and threads 
it through LLVM IR, MIR lowering, Clang builtins/tests, and MLIR ROCDL dialect 
so all layers agree on the new operand
- Updates AMDGPUWmmaIntrinsicModsAB so the clamp attribute is emitted, teaches 
VOP3P encoding to accept the immediate, and adjusts Clang codegen/builtin 
headers plus MLIR op definitions and tests to match

>From df40111bf084022085b4facd555c03fd42c2827d Mon Sep 17 00:00:00 2001
From: 0xzre 
Date: Mon, 8 Dec 2025 07:32:45 +0700
Subject: [PATCH] [AMDGPU] add clamp immediate operand to WMMA iu8 intrinsic

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def |  2 +-
 .../builtins-amdgcn-gfx1250-wmma-w32.cl  |  4 ++--
 ...iltins-amdgcn-error-gfx1250-wmma-w32-param.cl |  9 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td |  3 ++-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td  |  4 +++-
 .../UniformityAnalysis/AMDGPU/intrinsics.ll  |  6 +++---
 .../AMDGPU/llvm.amdgcn.wmma.gfx1250.w32.ll   |  4 ++--
 .../AMDGPU/llvm.amdgcn.wmma.imm.gfx1250.w32.ll   |  8 
 .../AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll  |  6 +++---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  5 +++--
 mlir/test/Target/LLVMIR/rocdl.mlir   | 16 ++--
 11 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 8af6ce1528a45..ebdac12ce107b 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -856,7 +856,7 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_fp8, 
"V8hV8iV8iIsV8hIbIb",
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_fp8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_fp8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x64_bf8_bf8, 
"V8hV8iV8iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
-TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, "V8iIbV8iIbV8iV8iIbIb", 
"nc", "gfx1250-insts,wavefrontsize32")
+TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x64_iu8, 
"V8iIbV8iIbV8iV8iIbIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_fp8_bf8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
 TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x128_bf8_fp8, 
"V8hV16iV16iIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl 
b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
index bdb1a7f0bb32f..41c2eb2155b89 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
@@ -148,13 +148,13 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c)
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_i32_16x16x64_iu8(
 // CHECK-GFX1250-NEXT:  entry:
-// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true)
+// CHECK-GFX1250-NEXT:[[TMP0:%.*]] = tail call <8 x i32> 
@llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 false, <8 x i32> [[A:%.*]], 
i1 false, <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false, i1 true, i1 false)
 // CHECK-GFX1250-NEXT:store <8 x i32> [[TMP0]], ptr addrspace(1) 
[[OUT:%.*]], align 32, !tbaa [[TBAA4]]
 // CHECK-GFX1250-NEXT:ret void
 //
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true);
+  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, 0, b, c, false, true, 
false);
 }
 
 // CHECK-GFX1250-LABEL: @test_amdgcn_wmma_f32_16x16x128_f8f6f4(
diff --git 
a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl 
b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
index 49ef2e571740c..8821524fde2db 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
@@ -108,10 +108,11 @@ void test_amdgcn_wmma_f16_16x16x64_bf8_bf8(global v8h* 
out, v8i a, v8i b, v8h c,
 
 void test_amdgcn_wmma_i32_16x16x64_iu8(global v8i* out, v8i a, v8i b, v8i c, 
int mod)
 {
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(mod, a, 0, b, c, false, 
false); // expected-error {{'__builtin_amdgcn_wmma_i32_16x16x64_iu8' must be a 
constant integer}}
-  *out = __builtin_amdgcn_wmma_i32_16x16x64_iu8(0, a, mod, b, c, false, 
false);

64 matches

Mail list logo