[AMDGPU] Move frame pointer from s34 to s33

Remove the gap left between the stack pointer (s32) and frame pointer
(s34) now that the scratch wave offset is no longer a part of the
calling convention ABI.

Update llvm/docs/AMDGPUUsage.rst to reflect the change.

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D75657

Author: slinder1

llvm/docs/AMDGPUUsage.rst

@@ -6067,7 +6067,7 @@ Frame Pointer
 +++++++++++++
 
 If the kernel needs a frame pointer for the reasons defined in
-``SIFrameLowering`` then SGPR34 is used and is always set to ``0`` in the
+``SIFrameLowering`` then SGPR33 is used and is always set to ``0`` in the
 kernel prolog. If a frame pointer is not required then all uses of the frame
 pointer are replaced with immediate ``0`` offsets.
 
@@ -8897,7 +8897,7 @@ registers and some in memory.
 The following is not part of the AMDGPU function calling convention but
 describes how the AMDGPU implements function calls:
 
-1.  SGPR34 is used as a frame pointer (FP) if necessary. Like the SP it is an
+1.  SGPR33 is used as a frame pointer (FP) if necessary. Like the SP it is an
     unswizzled scratch address. It is only needed if runtime sized ``alloca``
     are used, or for the reasons defined in ``SIFrameLowering``.
 2.  Runtime stack alignment is not currently supported.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -1950,7 +1950,7 @@ static void reservePrivateMemoryRegs(const TargetMachine &TM,
   // finalized, because it does not rely on the known stack size, only
   // properties like whether variable sized objects are present.
   if (ST.getFrameLowering()->hasFP(MF)) {
-    Info.setFrameOffsetReg(AMDGPU::SGPR34);
+    Info.setFrameOffsetReg(AMDGPU::SGPR33);
   }
 }
 

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

@@ -81,7 +81,7 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
     ScratchRSrcReg = AMDGPU::SGPR0_SGPR1_SGPR2_SGPR3;
 
     // TODO: Pick a high register, and shift down, similar to a kernel.
-    FrameOffsetReg = AMDGPU::SGPR34;
+    FrameOffsetReg = AMDGPU::SGPR33;
     StackPtrOffsetReg = AMDGPU::SGPR32;
 
     ArgInfo.PrivateSegmentBuffer =

llvm/test/CodeGen/AMDGPU/call-graph-register-usage.ll

@@ -13,14 +13,14 @@ define void @use_vcc() #1 {
 }
 
 ; GCN-LABEL: {{^}}indirect_use_vcc:
-; GCN: v_writelane_b32 v32, s34, 2
+; GCN: v_writelane_b32 v32, s33, 2
 ; GCN: v_writelane_b32 v32, s30, 0
 ; GCN: v_writelane_b32 v32, s31, 1
 ; GCN: s_swappc_b64
 ; GCN: v_readlane_b32 s4, v32, 0
 ; GCN: v_readlane_b32 s5, v32, 1
-; GCN: v_readlane_b32 s34, v32, 2
-; GCN: ; NumSgprs: 37
+; GCN: v_readlane_b32 s33, v32, 2
+; GCN: ; NumSgprs: 36
 ; GCN: ; NumVgprs: 33
 define void @indirect_use_vcc() #1 {
   call void @use_vcc()
@@ -29,8 +29,8 @@ define void @indirect_use_vcc() #1 {
 
 ; GCN-LABEL: {{^}}indirect_2level_use_vcc_kernel:
 ; GCN: is_dynamic_callstack = 0
-; CI: ; NumSgprs: 39
-; VI-NOBUG: ; NumSgprs: 41
+; CI: ; NumSgprs: 38
+; VI-NOBUG: ; NumSgprs: 40
 ; VI-BUG: ; NumSgprs: 96
 ; GCN: ; NumVgprs: 33
 define amdgpu_kernel void @indirect_2level_use_vcc_kernel(i32 addrspace(1)* %out) #0 {
@@ -48,8 +48,8 @@ define void @use_flat_scratch() #1 {
 }
 
 ; GCN-LABEL: {{^}}indirect_use_flat_scratch:
-; CI: ; NumSgprs: 39
-; VI: ; NumSgprs: 41
+; CI: ; NumSgprs: 38
+; VI: ; NumSgprs: 40
 ; GCN: ; NumVgprs: 33
 define void @indirect_use_flat_scratch() #1 {
   call void @use_flat_scratch()
@@ -58,8 +58,8 @@ define void @indirect_use_flat_scratch() #1 {
 
 ; GCN-LABEL: {{^}}indirect_2level_use_flat_scratch_kernel:
 ; GCN: is_dynamic_callstack = 0
-; CI: ; NumSgprs: 39
-; VI-NOBUG: ; NumSgprs: 41
+; CI: ; NumSgprs: 38
+; VI-NOBUG: ; NumSgprs: 40
 ; VI-BUG: ; NumSgprs: 96
 ; GCN: ; NumVgprs: 33
 define amdgpu_kernel void @indirect_2level_use_flat_scratch_kernel(i32 addrspace(1)* %out) #0 {

llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll

@@ -23,9 +23,9 @@ define amdgpu_kernel void @test_kernel_call_external_void_func_void_clobber_s30_
 
 ; GCN-LABEL: {{^}}test_func_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void:
 ; GCN: buffer_store_dword
-; GCN: v_writelane_b32 v32, s34, 4
-; GCN: v_writelane_b32 v32, s36, 0
-; GCN: v_writelane_b32 v32, s37, 1
+; GCN: v_writelane_b32 v32, s33, 4
+; GCN: v_writelane_b32 v32, s34, 0
+; GCN: v_writelane_b32 v32, s35, 1
 ; GCN: v_writelane_b32 v32, s30, 2
 ; GCN: v_writelane_b32 v32, s31, 3
 
@@ -35,10 +35,10 @@ define amdgpu_kernel void @test_kernel_call_external_void_func_void_clobber_s30_
 ; GCN-NEXT: s_swappc_b64
 ; GCN-DAG: v_readlane_b32 s4, v32, 2
 ; GCN-DAG: v_readlane_b32 s5, v32, 3
-; GCN: v_readlane_b32 s37, v32, 1
-; GCN: v_readlane_b32 s36, v32, 0
+; GCN: v_readlane_b32 s35, v32, 1
+; GCN: v_readlane_b32 s34, v32, 0
 
-; GCN: v_readlane_b32 s34, v32, 4
+; GCN: v_readlane_b32 s33, v32, 4
 ; GCN: buffer_load_dword
 ; GCN: s_setpc_b64
 define void @test_func_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void() #0 {
@@ -50,14 +50,14 @@ define void @test_func_call_external_void_func_void_clobber_s30_s31_call_externa
 
 ; GCN-LABEL: {{^}}test_func_call_external_void_funcx2:
 ; GCN: buffer_store_dword v32
-; GCN: v_writelane_b32 v32, s34, 4
+; GCN: v_writelane_b32 v32, s33, 4
 
-; GCN: s_mov_b32 s34, s32
+; GCN: s_mov_b32 s33, s32
 ; GCN: s_add_u32 s32, s32, 0x400
 ; GCN: s_swappc_b64
 ; GCN-NEXT: s_swappc_b64
 
-; GCN: v_readlane_b32 s34, v32, 4
+; GCN: v_readlane_b32 s33, v32, 4
 ; GCN: buffer_load_dword v32,
 define void @test_func_call_external_void_funcx2() #0 {
   call void @external_void_func_void()
@@ -125,6 +125,8 @@ define amdgpu_kernel void @test_call_void_func_void_mayclobber_v31(i32 addrspace
   ret void
 }
 
+; FIXME: What is the expected behavior for reserved registers here?
+
 ; GCN-LABEL: {{^}}test_call_void_func_void_preserves_s33:
 ; GCN: s_getpc_b64 s[4:5]
 ; GCN-NEXT: s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
@@ -146,8 +148,6 @@ define amdgpu_kernel void @test_call_void_func_void_preserves_s33(i32 addrspace(
   ret void
 }
 
-; FIXME: What is the expected behavior for reserved registers here?
-
 ; GCN-LABEL: {{^}}test_call_void_func_void_preserves_s34: {{.*}}
 ; GCN-NOT: s34