AMDGPU: Directly annotate functions if they have calls

Currently we infer whether the flat-scratch-init kernel input should
be enabled based on calls. Move this handling, so we can decide if the
full set of ABI inputs is needed in kernels. Ideally we would have an
analysis of some sort, rather than the function attributes.

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
index 66801b7..ab55ee4 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
@@ -277,7 +277,6 @@ bool AMDGPUAnnotateKernelFeatures::propagateUniformWorkGroupAttribute(
 
 bool AMDGPUAnnotateKernelFeatures::addFeatureAttributes(Function &F) {
   const GCNSubtarget &ST = TM->getSubtarget<GCNSubtarget>(F);
-  bool HasFlat = ST.hasFlatAddressSpace();
   bool HasApertureRegs = ST.hasApertureRegs();
   SmallPtrSet<const Constant *, 8> ConstantExprVisited;
 
@@ -347,8 +346,8 @@ bool AMDGPUAnnotateKernelFeatures::addFeatureAttributes(Function &F) {
   // TODO: We could refine this to captured pointers that could possibly be
   // accessed by flat instructions. For now this is mostly a poor way of
   // estimating whether there are calls before argument lowering.
-  if (HasFlat && !IsFunc && HaveCall) {
-    F.addFnAttr("amdgpu-flat-scratch");
+  if (!IsFunc && HaveCall) {
+    F.addFnAttr("amdgpu-calls");
     Changed = true;
   }
 

diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 0c67b14..300b288 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -54,6 +54,11 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
 
   Occupancy = ST.computeOccupancy(MF, getLDSSize());
   CallingConv::ID CC = F.getCallingConv();
+  const MachineFrameInfo &FrameInfo = MF.getFrameInfo();
+
+  // FIXME: Should have analysis or something rather than attribute to detect
+  // calls.
+  const bool HasCalls = FrameInfo.hasCalls() || F.hasFnAttribute("amdgpu-calls");
 
   if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL) {
     if (!F.arg_empty())
@@ -107,7 +112,6 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
   if (F.hasFnAttribute("amdgpu-work-item-id-z"))
     WorkItemIDZ = true;
 
-  const MachineFrameInfo &FrameInfo = MF.getFrameInfo();
   bool HasStackObjects = FrameInfo.hasStackObjects();
 
   if (isEntryFunction()) {
@@ -158,7 +162,7 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
     };
     // TODO: This could be refined a lot. The attribute is a poor way of
     // detecting calls that may require it before argument lowering.
-    if (hasNonSpillStackObjects() || F.hasFnAttribute("amdgpu-flat-scratch"))
+    if (HasCalls || hasNonSpillStackObjects())
       FlatScratchInit = true;
   }
 

diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
index ca6739f..82ba285 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
+++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
@@ -319,4 +319,4 @@ attributes #3 = { nounwind }
 ; HSA: attributes #16 = { nounwind "amdgpu-implicitarg-ptr" "target-cpu"="fiji" "uniform-work-group-size"="false" }
 ; HSA: attributes #17 = { nounwind "uniform-work-group-size"="false" }
 ; HSA: attributes #18 = { nounwind }
-; HSA: attributes #19 = { nounwind "amdgpu-flat-scratch" "uniform-work-group-size"="false" }
+; HSA: attributes #19 = { nounwind "amdgpu-calls" "uniform-work-group-size"="false" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll
index 51002e8..022ff85 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s 
+; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s
 
 ; If the kernel does not have the uniform-work-group-attribute, set both callee and caller as false
 
@@ -7,7 +7,7 @@ define void @foo() #0 {
   ret void
 }
 
-; CHECK: define amdgpu_kernel void @kernel1() #[[FOO]] {
+; CHECK: define amdgpu_kernel void @kernel1() #[[KERNEL1:[0-9]+]] {
 define amdgpu_kernel void @kernel1() #1 {
   call void @foo()
   ret void
@@ -16,3 +16,4 @@ define amdgpu_kernel void @kernel1() #1 {
 attributes #0 = { "uniform-work-group-size"="true" }
 
 ; CHECK: attributes #[[FOO]] = { "uniform-work-group-size"="false" }
+; CHECK: attributes #[[KERNEL1]] = { "amdgpu-calls" "uniform-work-group-size"="false" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll
index 8376114..321b8cb 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s 
+; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s
 
 ; Test to verify if the attribute gets propagated across nested function calls
 
@@ -13,7 +13,7 @@ define void @func2() #1 {
   ret void
 }
 
-; CHECK: define amdgpu_kernel void @kernel3() #[[FUNC:[0-9]+]] {
+; CHECK: define amdgpu_kernel void @kernel3() #[[KERNEL:[0-9]+]] {
 define amdgpu_kernel void @kernel3() #2 {
   call void @func2()
   ret void
@@ -22,3 +22,4 @@ define amdgpu_kernel void @kernel3() #2 {
 attributes #2 = { "uniform-work-group-size"="true" }
 
 ; CHECK: attributes #[[FUNC]] = { "uniform-work-group-size"="true" }
+; CHECK: attributes #[[KERNEL]] = { "amdgpu-calls" "uniform-work-group-size"="true" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll
index 4a332f6..78136d9 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s 
+; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s
 
 ; Two kernels with different values of the uniform-work-group-attribute call the same function
 
@@ -13,7 +13,7 @@ define amdgpu_kernel void @kernel1() #1 {
   ret void
 }
 
-; CHECK: define amdgpu_kernel void @kernel2() #[[FUNC]] {
+; CHECK: define amdgpu_kernel void @kernel2() #[[KERNEL2:[0-9]+]] {
 define amdgpu_kernel void @kernel2() #2 {
   call void @func()
   ret void
@@ -22,4 +22,5 @@ define amdgpu_kernel void @kernel2() #2 {
 attributes #1 = { "uniform-work-group-size"="true" }
 
 ; CHECK: attributes #[[FUNC]] = { "uniform-work-group-size"="false" }
-; CHECK: attributes #[[KERNEL1]] = { "uniform-work-group-size"="true" }
+; CHECK: attributes #[[KERNEL1]] = { "amdgpu-calls" "uniform-work-group-size"="true" }
+; CHECK: attributes #[[KERNEL2]] = { "amdgpu-calls" "uniform-work-group-size"="false" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll
index 15131a4..4214587 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s 
+; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s
 
 ; Propagate the uniform-work-group-attribute from the kernel to callee if it doesn't have it
 ; CHECK: define void @func() #[[FUNC:[0-9]+]] {
@@ -29,5 +29,5 @@ attributes #1 = { "uniform-work-group-size"="false" }
 attributes #2 = { "uniform-work-group-size"="true" }
 
 ; CHECK: attributes #[[FUNC]] = { nounwind "uniform-work-group-size"="false" }
-; CHECK: attributes #[[KERNEL1]] = { "uniform-work-group-size"="false" }
-; CHECK: attributes #[[KERNEL2]] = { "uniform-work-group-size"="true" }
+; CHECK: attributes #[[KERNEL1]] = { "amdgpu-calls" "uniform-work-group-size"="false" }
+; CHECK: attributes #[[KERNEL2]] = { "amdgpu-calls" "uniform-work-group-size"="true" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll
index 9d07a88..0b6053f 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s 
+; RUN: opt -S -mtriple=amdgcn-amd- -amdgpu-annotate-kernel-features %s | FileCheck %s
 
 ; Test to ensure recursive functions exhibit proper behaviour
 ; Test to generate fibonacci numbers
@@ -25,7 +25,7 @@ exit:
   ret i32 1
 }
 
-; CHECK: define amdgpu_kernel void @kernel(i32 addrspace(1)* %m) #[[FIB]] {
+; CHECK: define amdgpu_kernel void @kernel(i32 addrspace(1)* %m) #[[KERNEL:[0-9]+]] {
 define amdgpu_kernel void @kernel(i32 addrspace(1)* %m) #1 {
   %r = call i32 @fib(i32 5)
   store i32 %r, i32 addrspace(1)* %m
@@ -35,3 +35,4 @@ define amdgpu_kernel void @kernel(i32 addrspace(1)* %m) #1 {
 attributes #1 = { "uniform-work-group-size"="true" }
 
 ; CHECK: attributes #[[FIB]] = { "uniform-work-group-size"="true" }
+; CHECK: attributes #[[KERNEL]] = { "amdgpu-calls" "uniform-work-group-size"="true" }

diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll
index 0a3cae8..a9a2f12 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll
+++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll
@@ -32,4 +32,4 @@ define amdgpu_kernel void @kernel3() #0 {
 
 attributes #0 = { "uniform-work-group-size"="false" }
 
-; CHECK: attributes #[[FUNC]] = { "amdgpu-flat-scratch" "uniform-work-group-size"="false" }
+; CHECK: attributes #[[FUNC]] = { "amdgpu-calls" "uniform-work-group-size"="false" }