[AMDGPU][SplitModule] Allow non-kernels to be treated as roots (llvm#95902)

I initially assumed only kernels could be roots, but that is wrong. A
function with no callers also needs to be a root to ensure it is
correctly handled.
They're very rare because we usually internalize everything, and
internal functions with no callers would be deleted.

When they are present, we need to also consider their dependencies and
act accordingly. Previously, we could put a function "by default" in P0,
but it could call another function with internal linkage defined in
another module which was of course incorrect.

Fixes SWDEV-467695

Author: Pierre-vh

llvm/lib/Target/AMDGPU/AMDGPUSplitModule.cpp

@@ -1,4 +1,4 @@
-; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-kernel-threshold=0
+; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-function-threshold=0
 ; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 %s
 ; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 %s
 ; RUN: llvm-dis -o - %t2 | FileCheck --check-prefix=CHECK2 %s

llvm/test/tools/llvm-split/AMDGPU/address-taken-externalize-with-call.ll

@@ -1,4 +1,4 @@
-; RUN: llvm-split -o %t %s -j 2 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-kernel-threshold=0
+; RUN: llvm-split -o %t %s -j 2 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-function-threshold=0
 ; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 %s
 ; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 %s
 

llvm/test/tools/llvm-split/AMDGPU/address-taken-externalize.ll

@@ -0,0 +1,36 @@
+; RUN: llvm-split -o %t %s -j 2 -mtriple amdgcn-amd-amdhsa -debug 2>&1 | FileCheck %s --implicit-check-not="[root]"
+; REQUIRES: asserts
+
+; func_3 is never directly called, it needs to be considered
+; as a root to handle this module correctly.
+
+; CHECK:      [root] kernel_1
+; CHECK-NEXT:   [dependency] func_1
+; CHECK-NEXT:   [dependency] func_2
+; CHECK-NEXT: [root] func_3
+; CHECK-NEXT:   [dependency] func_2
+
+define amdgpu_kernel void @kernel_1() {
+entry:
+  call void @func_1()
+  ret void
+}
+
+define linkonce_odr hidden void @func_1() {
+entry:
+  %call = call i32 @func_2()
+  ret void
+}
+
+define linkonce_odr hidden i32 @func_2() #0 {
+entry:
+  ret i32 0
+}
+
+define void @func_3() {
+entry:
+  %call = call i32 @func_2()
+  ret void
+}
+
+attributes #0 = { noinline optnone }

llvm/test/tools/llvm-split/AMDGPU/debug-non-kernel-root.ll

@@ -1,9 +1,9 @@
-; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-kernel-threshold=1.2 -amdgpu-module-splitting-large-kernel-merge-overlap=0.5
+; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-function-threshold=1.2 -amdgpu-module-splitting-large-function-merge-overlap=0.5
 ; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 %s
 ; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 %s
 ; RUN: llvm-dis -o - %t2 | FileCheck --check-prefix=CHECK2 %s
 
-; RUN: llvm-split -o %t.nolarge %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-kernel-threshold=0
+; RUN: llvm-split -o %t.nolarge %s -j 3 -mtriple amdgcn-amd-amdhsa -amdgpu-module-splitting-large-function-threshold=0
 ; RUN: llvm-dis -o - %t.nolarge0 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK0 %s
 ; RUN: llvm-dis -o - %t.nolarge1 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK1 %s
 ; RUN: llvm-dis -o - %t.nolarge2 | FileCheck --check-prefix=NOLARGEKERNELS-CHECK2 %s

llvm/test/tools/llvm-split/AMDGPU/large-kernels-merging.ll

@@ -0,0 +1,44 @@
+; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa
+; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 --implicit-check-not=DEFINE %s
+; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 --implicit-check-not=DEFINE %s
+; RUN: llvm-dis -o - %t2 | FileCheck --check-prefix=CHECK2 --implicit-check-not=DEFINE %s
+
+; 3 functions with each their own dependencies should go into 3
+; distinct partitions.
+
+; CHECK0: define void @C
+; CHECK0: define internal void @HelperC
+
+; CHECK1: define void @B
+; CHECK1: define internal void @HelperB
+
+; CHECK2: define void @A
+; CHECK2: define internal void @HelperA
+
+
+define void @A() {
+  call void @HelperA()
+  ret void
+}
+
+define internal void @HelperA() {
+  ret void
+}
+
+define void @B() {
+  call void @HelperB()
+  ret void
+}
+
+define internal void @HelperB() {
+  ret void
+}
+
+define void @C() {
+  call void @HelperC()
+  ret void
+}
+
+define internal void @HelperC() {
+  ret void
+}

llvm/test/tools/llvm-split/AMDGPU/non-kernels-dependencies.ll

@@ -0,0 +1,72 @@
+; RUN: llvm-split -o %t %s -j 3 -mtriple amdgcn-amd-amdhsa
+; RUN: llvm-dis -o - %t0 | FileCheck --check-prefix=CHECK0 --implicit-check-not=DEFINE %s
+; RUN: llvm-dis -o - %t1 | FileCheck --check-prefix=CHECK1 --implicit-check-not=DEFINE %s
+; RUN: llvm-dis -o - %t2 | FileCheck --check-prefix=CHECK2 --implicit-check-not=DEFINE %s
+
+; We have 4 function:
+;   - Each function has an internal helper
+;   - @A and @B's helpers does an indirect call.
+;
+; For non-kernels, indirect calls shouldn't matter, so
+; @CallCandidate doesn't have to be in A/B's partition, unlike
+; in the corresponding tests for kernels where it has to.
+
+; CHECK0: define hidden void @HelperA
+; CHECK0: define hidden void @HelperB
+; CHECK0: define internal void @HelperC
+; CHECK0: define internal void @HelperD
+; CHECK0: define void @A
+; CHECK0: define void @B
+
+; CHECK1: define internal void @HelperD
+; CHECK1: define void @D
+
+; CHECK2: define hidden void @CallCandidate
+; CHECK2: define internal void @HelperC
+; CHECK2: define void @C
+
+@addrthief = global [3 x ptr] [ptr @HelperA, ptr @HelperB, ptr @CallCandidate]
+
+define internal void @HelperA(ptr %call) {
+  call void %call()
+  ret void
+}
+
+define internal void @HelperB(ptr %call) {
+  call void @HelperC()
+  call void %call()
+  call void @HelperD()
+  ret void
+}
+
+define internal void @CallCandidate() {
+  ret void
+}
+
+define internal void @HelperC() {
+  ret void
+}
+
+define internal void @HelperD() {
+  ret void
+}
+
+define void @A(ptr %call) {
+  call void @HelperA(ptr %call)
+  ret void
+}
+
+define void @B(ptr %call) {
+  call void @HelperB(ptr %call)
+  ret void
+}
+
+define void @C() {
+  call void @HelperC()
+  ret void
+}
+
+define void @D() {
+  call void @HelperD()
+  ret void
+}