From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Wed, 27 Mar 2019 17:14:32 +0000 (+0000)
Subject: AMDGPU: Enable the scavenger for large frames
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=0755a8d19c51909534cca7b02e1db02421998049;p=llvm

AMDGPU: Enable the scavenger for large frames

Another test is needed for the case where the scavenge fail, but
there's another issue with that which needs an additional fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@357093 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/SIRegisterInfo.cpp b/lib/Target/AMDGPU/SIRegisterInfo.cpp
index 8e49b647aa7..92fbbcdc93d 100644
--- a/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -257,11 +257,20 @@ bool SIRegisterInfo::requiresFrameIndexScavenging(
 
 bool SIRegisterInfo::requiresFrameIndexReplacementScavenging(
   const MachineFunction &MF) const {
-  // m0 is needed for the scalar store offset. m0 is unallocatable, so we can't
-  // create a virtual register for it during frame index elimination, so the
-  // scavenger is directly needed.
-  return MF.getFrameInfo().hasStackObjects() &&
-         MF.getSubtarget<GCNSubtarget>().hasScalarStores() &&
+  const MachineFrameInfo &MFI = MF.getFrameInfo();
+  if (!MFI.hasStackObjects())
+    return false;
+
+  // The scavenger is used for large frames which may require finding a free
+  // register for large offsets.
+  if (!isUInt<12>(MFI.getStackSize()))
+    return true;
+
+  // If using scalar stores, for spills, m0 is needed for the scalar store
+  // offset (pre-GFX9). m0 is unallocatable, so we can't create a virtual
+  // register for it during frame index elimination, so the scavenger is
+  // directly needed.
+  return MF.getSubtarget<GCNSubtarget>().hasScalarStores() &&
          MF.getInfo<SIMachineFunctionInfo>()->hasSpilledSGPRs();
 }
 
diff --git a/test/CodeGen/AMDGPU/spill-offset-calculation.ll b/test/CodeGen/AMDGPU/spill-offset-calculation.ll
index 3c179b580fc..14cbf5518e1 100644
--- a/test/CodeGen/AMDGPU/spill-offset-calculation.ll
+++ b/test/CodeGen/AMDGPU/spill-offset-calculation.ll
@@ -35,9 +35,8 @@ entry:
 
   %aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
   ; 0x40000 / 64 = 4096 (for wave64)
-  ; CHECK: s_add_u32 s7, s7, 0x40000
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s7 ; 4-byte Folded Spill
-  ; CHECK: s_sub_u32 s7, s7, 0x40000
+  ; CHECK: s_add_u32 s6, s7, 0x40000
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
   %a = load volatile i32, i32 addrspace(5)* %aptr
 
   ; Force %a to spill
@@ -88,10 +87,9 @@ entry:
   %bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
 
   ; 0x3ff00 / 64 = 4092 (for wave64)
-  ; CHECK: s_add_u32 s7, s7, 0x3ff00
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s7 ; 4-byte Folded Spill
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s7 offset:4 ; 4-byte Folded Spill
-  ; CHECK: s_sub_u32 s7, s7, 0x3ff00
+  ; CHECK: s_add_u32 s6, s7, 0x3ff00
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 offset:4 ; 4-byte Folded Spill
   %aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
   %a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr
 
@@ -139,9 +137,8 @@ entry:
 
   %aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
   ; 0x40000 / 64 = 4096 (for wave64)
-  ; CHECK: s_add_u32 s5, s5, 0x40000
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s5 ; 4-byte Folded Spill
-  ; CHECK: s_sub_u32 s5, s5, 0x40000
+  ; CHECK: s_add_u32 s6, s5, 0x40000
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
   %a = load volatile i32, i32 addrspace(5)* %aptr
 
   ; Force %a to spill
@@ -192,10 +189,9 @@ entry:
   %bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
 
   ; 0x3ff00 / 64 = 4092 (for wave64)
-  ; CHECK: s_add_u32 s5, s5, 0x3ff00
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s5 ; 4-byte Folded Spill
-  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s5 offset:4 ; 4-byte Folded Spill
-  ; CHECK: s_sub_u32 s5, s5, 0x3ff00
+  ; CHECK: s_add_u32 s6, s5, 0x3ff00
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
+  ; CHECK: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 offset:4 ; 4-byte Folded Spill
   %aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
   %a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr