[NVPTX] Fix internal indirect call prototypes not obeying the ABI #100131
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-libc @llvm/pr-subscribers-backend-nvptx Author: Joseph Huber (jhuber6) ChangesSummary: Fixes #100055 Full diff: https://github.com/llvm/llvm-project/pull/100131.diff 3 Files Affected:
diff --git a/libc/config/gpu/entrypoints.txt b/libc/config/gpu/entrypoints.txt
index 42909cec55890..fa878d8999227 100644
--- a/libc/config/gpu/entrypoints.txt
+++ b/libc/config/gpu/entrypoints.txt
@@ -1,13 +1,3 @@
-if(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
- set(extra_entrypoints
- # stdio.h entrypoints
- libc.src.stdio.snprintf
- libc.src.stdio.sprintf
- libc.src.stdio.vsnprintf
- libc.src.stdio.vsprintf
- )
-endif()
-
set(TARGET_LIBC_ENTRYPOINTS
# assert.h entrypoints
libc.src.assert.__assert_fail
@@ -186,13 +176,16 @@ set(TARGET_LIBC_ENTRYPOINTS
libc.src.errno.errno
# stdio.h entrypoints
- ${extra_entrypoints}
libc.src.stdio.clearerr
libc.src.stdio.fclose
libc.src.stdio.printf
libc.src.stdio.vprintf
libc.src.stdio.fprintf
libc.src.stdio.vfprintf
+ libc.src.stdio.snprintf
+ libc.src.stdio.sprintf
+ libc.src.stdio.vsnprintf
+ libc.src.stdio.vsprintf
libc.src.stdio.feof
libc.src.stdio.ferror
libc.src.stdio.fflush
diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
index 44c1a2e50486c..bcd2ee2f4fc1e 100644
--- a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -1429,7 +1429,6 @@ std::string NVPTXTargetLowering::getPrototype(
bool first = true;
- const Function *F = CB.getFunction();
unsigned NumArgs = VAInfo ? VAInfo->first : Args.size();
for (unsigned i = 0, OIdx = 0; i != NumArgs; ++i, ++OIdx) {
Type *Ty = Args[i].Ty;
@@ -1471,10 +1470,11 @@ std::string NVPTXTargetLowering::getPrototype(
continue;
}
+ // Indirect calls need strict ABI alignment so we disable optimizations.
Type *ETy = Args[i].IndirectType;
Align InitialAlign = Outs[OIdx].Flags.getNonZeroByValAlign();
Align ParamByValAlign =
- getFunctionByValParamAlign(F, ETy, InitialAlign, DL);
+ getFunctionByValParamAlign(/*F=*/nullptr, ETy, InitialAlign, DL);
O << ".param .align " << ParamByValAlign.value() << " .b8 ";
O << "_";
diff --git a/llvm/test/CodeGen/NVPTX/indirect_byval.ll b/llvm/test/CodeGen/NVPTX/indirect_byval.ll
new file mode 100644
index 0000000000000..ac6c4e262fd60
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/indirect_byval.ll
@@ -0,0 +1,94 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_52 -mattr=+ptx64 | FileCheck %s
+; RUN: %if ptxas %{ llc < %s -march=nvptx64 -mcpu=sm_52 -mattr=+ptx64 | %ptxas-verify %}
+
+target triple = "nvptx64-nvidia-cuda"
+
+%struct.S = type { i8 }
+%struct.U = type { i64 }
+
+@ptr = external global ptr, align 8
+
+define internal i32 @foo() {
+; CHECK-LABEL: foo(
+; CHECK: {
+; CHECK-NEXT: .local .align 1 .b8 __local_depot0[2];
+; CHECK-NEXT: .reg .b64 %SP;
+; CHECK-NEXT: .reg .b64 %SPL;
+; CHECK-NEXT: .reg .b16 %rs<2>;
+; CHECK-NEXT: .reg .b32 %r<3>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0: // %entry
+; CHECK-NEXT: mov.u64 %SPL, __local_depot0;
+; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
+; CHECK-NEXT: ld.global.u64 %rd1, [ptr];
+; CHECK-NEXT: ld.u8 %rs1, [%SP+1];
+; CHECK-NEXT: add.u64 %rd2, %SP, 0;
+; CHECK-NEXT: { // callseq 0, 0
+; CHECK-NEXT: .param .align 1 .b8 param0[1];
+; CHECK-NEXT: st.param.b8 [param0+0], %rs1;
+; CHECK-NEXT: .param .b64 param1;
+; CHECK-NEXT: st.param.b64 [param1+0], %rd2;
+; CHECK-NEXT: .param .b32 retval0;
+; CHECK-NEXT: prototype_0 : .callprototype (.param .b32 _) _ (.param .align 1 .b8 _[1], .param .b64 _);
+; CHECK-NEXT: call (retval0),
+; CHECK-NEXT: %rd1,
+; CHECK-NEXT: (
+; CHECK-NEXT: param0,
+; CHECK-NEXT: param1
+; CHECK-NEXT: )
+; CHECK-NEXT: , prototype_0;
+; CHECK-NEXT: ld.param.b32 %r1, [retval0+0];
+; CHECK-NEXT: } // callseq 0
+; CHECK-NEXT: st.param.b32 [func_retval0+0], %r1;
+; CHECK-NEXT: ret;
+entry:
+ %s = alloca %struct.S, align 1
+ %agg.tmp = alloca %struct.S, align 1
+ %0 = load ptr, ptr @ptr, align 8
+ %call = call i32 %0(ptr byval(%struct.S) align 1 %agg.tmp, ptr noundef %s)
+ ret i32 %call
+}
+
+define internal i32 @bar() {
+; CHECK-LABEL: bar(
+; CHECK: // @bar
+; CHECK-NEXT: {
+; CHECK-NEXT: .local .align 8 .b8 __local_depot1[16];
+; CHECK-NEXT: .reg .b64 %SP;
+; CHECK-NEXT: .reg .b64 %SPL;
+; CHECK-NEXT: .reg .b32 %r<3>;
+; CHECK-NEXT: .reg .b64 %rd<4>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0: // %entry
+; CHECK-NEXT: mov.u64 %SPL, __local_depot1;
+; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
+; CHECK-NEXT: ld.global.u64 %rd1, [ptr];
+; CHECK-NEXT: ld.u64 %rd2, [%SP+8];
+; CHECK-NEXT: add.u64 %rd3, %SP, 0;
+; CHECK-NEXT: { // callseq 1, 0
+; CHECK-NEXT: .param .align 8 .b8 param0[8];
+; CHECK-NEXT: st.param.b64 [param0+0], %rd2;
+; CHECK-NEXT: .param .b64 param1;
+; CHECK-NEXT: st.param.b64 [param1+0], %rd3;
+; CHECK-NEXT: .param .b32 retval0;
+; CHECK-NEXT: prototype_1 : .callprototype (.param .b32 _) _ (.param .align 8 .b8 _[8], .param .b64 _);
+; CHECK-NEXT: call (retval0),
+; CHECK-NEXT: %rd1,
+; CHECK-NEXT: (
+; CHECK-NEXT: param0,
+; CHECK-NEXT: param1
+; CHECK-NEXT: )
+; CHECK-NEXT: , prototype_1;
+; CHECK-NEXT: ld.param.b32 %r1, [retval0+0];
+; CHECK-NEXT: } // callseq 1
+; CHECK-NEXT: st.param.b32 [func_retval0+0], %r1;
+; CHECK-NEXT: ret;
+entry:
+ %s = alloca %struct.U, align 8
+ %agg.tmp = alloca %struct.U, align 8
+ %0 = load ptr, ptr @ptr, align 8
+ %call = call noundef i32 %0(ptr byval(%struct.U) align 8 %agg.tmp, ptr %s)
+ ret i32 %call
+}
|
|
The |
Artem-B
approved these changes
Jul 23, 2024
| Type *ETy = Args[i].IndirectType; | ||
| Align InitialAlign = Outs[OIdx].Flags.getNonZeroByValAlign(); | ||
| Align ParamByValAlign = | ||
| getFunctionByValParamAlign(F, ETy, InitialAlign, DL); | ||
| getFunctionByValParamAlign(/*F=*/nullptr, ETy, InitialAlign, DL); |
There was a problem hiding this comment.
The comment above needs to mantion that it's the nullptr here that forces strict ABI-compliant alignment. Otherwise it's not clear which part of the code controls this.
Ideally, it would be even nicer to make "ABI-compliant alignment" a parameter.
There was a problem hiding this comment.
This doesn't really make sense to me as a fix. Indirect calls shouldn't require special handling vs. direct
There was a problem hiding this comment.
The NVPTX backend thinks that internal functions can have their ABI modified, so they up-scale alignment on things. This doesn't work for these indirect calls, which seem to strictly require the ABI.
Summary: The NVPTX backend optimizes the ABI for functions that are internal, however, this is not legal for indirect call prototypes. Previously, we would modify the ABI on an aggregate byval type passed to an indirect call prototype, which would make PTXAS error. This patch just passes the function as a nullptr to force strict ABI compliance without modification in the helper function. Fixes llvm#100055
jhuber6
force-pushed
the
FixNVPTXCallproto
branch
from
cbfdca3 to
1a779ff
Compare
llvmbot
pushed a commit
to llvmbot/llvm-project
that referenced
this pull request
Jul 23, 2024
…vm#100131) Summary: The NVPTX backend optimizes the ABI for functions that are internal, however, this is not legal for indirect call prototypes. Previously, we would modify the ABI on an aggregate byval type passed to an indirect call prototype, which would make PTXAS error. This patch just passes the function as a nullptr to force strict ABI compliance without modification in the helper function. Fixes llvm#100055 (cherry picked from commit e0649a5)
tru
pushed a commit
to llvmbot/llvm-project
that referenced
this pull request
Jul 24, 2024
…vm#100131) Summary: The NVPTX backend optimizes the ABI for functions that are internal, however, this is not legal for indirect call prototypes. Previously, we would modify the ABI on an aggregate byval type passed to an indirect call prototype, which would make PTXAS error. This patch just passes the function as a nullptr to force strict ABI compliance without modification in the helper function. Fixes llvm#100055 (cherry picked from commit e0649a5)
yuxuanchen1997
pushed a commit
that referenced
this pull request
Jul 25, 2024
…00131) Summary: The NVPTX backend optimizes the ABI for functions that are internal, however, this is not legal for indirect call prototypes. Previously, we would modify the ABI on an aggregate byval type passed to an indirect call prototype, which would make PTXAS error. This patch just passes the function as a nullptr to force strict ABI compliance without modification in the helper function. Fixes #100055 Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60251110
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
The NVPTX backend optimizes the ABI for functions that are internal,
however, this is not legal for indirect call prototypes. Previously, we
would modify the ABI on an aggregate byval type passed to an indirect
call prototype, which would make PTXAS error. This patch just passes the
function as a nullptr to force strict ABI compliance without
modification in the helper function.
Fixes #100055