[AMDGPU] Introduce more scratch registers in the ABI.

[cdevadas]

Summary

The AMDGPU target has a convention that defined all VGPRs
(except the initial 32 argument registers) as callee-saved.
This convention is not efficient always, esp. when the callee
requiring more registers, ended up emitting a large number of
spills, even though its caller requires only a few.

This patch revises the ABI by introducing more scratch registers
that a callee can freely use.
The 256 VGPR registers now become:

32 argument registers
112 scratch registers and
112 callee-saved registers.

The scratch registers and the CSRs are intermixed at regular
intervals (a split boundary of 8) to obtain a better occupancy.

[arsenm]

llvm/lib/Target/AMDGPU/SIRegisterInfo.td

220–223 ↗

(On Diff #251047)

This part should be split into a separate change

[cdevadas]

llvm/lib/Target/AMDGPU/SIRegisterInfo.td

220–223 ↗

(On Diff #251047)

Do you prefer it as a separate commit?

[arsenm]

llvm/lib/Target/AMDGPU/SIRegisterInfo.td

220–223 ↗

(On Diff #251047)

Yes, this can be committed independently

[cdevadas]

Reverted the changes made on the CostPerUse value.
It will go in a follow-up commit after this review.

[arsenm]

In D76356#1929542, @cdevadas wrote:

Reverted the changes made on the CostPerUse value.
It will go in a follow-up commit after this review.

I would expect this to be the preliminary commit

[cdevadas]

rebase

[arsenm]

llvm/docs/AMDGPUUsage.rst

6511

A description of why it's split this way may be helpful

[t-tye]

llvm/docs/AMDGPUUsage.rst

6511

Is the striping being picked at 4 VGPRs to match the hardware VGPR allocation granularity (4 for <=GFX9 and 8 for >=GFX10)? How does this stripping impact register file fragmentation? What is the impact of objects being promoted to registers that are larger than 4 VGPRs?

[arsenm]

llvm/docs/AMDGPUUsage.rst

6511

These aren't used for argument passing, so there's no concept of objects to consider

[t-tye]

llvm/docs/AMDGPUUsage.rst

6511

Also this is an ABI breaking change (as is the change for the handling of the wave scratch offset) so should the EI_ABIVERSION for each EI_OSABI in the ELF header be bumped? My thinking is no since AMD has not yet started to support isa level linking nor function pointers so this change cannot affect any existing programs.

[arsenm]

llvm/docs/AMDGPUUsage.rst

6511

We didn't have, and still don't, have an ABI worthy of considering for this

[b-sumner]

llvm/docs/AMDGPUUsage.rst

6511

Seems OK to me to not bump the version since we're not touching the kernel ABI and we don't have ISA linking.

[cdevadas]

Added a brief description of the register split (AMDGPUUsage.rst).

[arsenm]

llvm/docs/AMDGPUUsage.rst

6537

guarantee is a bit too strong of a claim

[t-tye]

llvm/docs/AMDGPUUsage.rst

6511

These aren't used for argument passing, so there's no concept of objects to consider

But if the compiler wants to promote an object to contiguous registers, then it will end up spanning both clobbered and non-clobbered registers for objects larger than the granularity, forcing some spill/restore code for the parts that are in non-clobbered registers. If the stripping was at a courser granularity then this may be unnecessary if the object will fit in the courser granularity.

Do we think this could be an issue?

[cdevadas]

llvm/docs/AMDGPUUsage.rst

6511

If you are still talking about passing the object, the compiler won't promote an object of large size into registers beyond the range defined by the convention (the first 32 VGPRs, in our case, that we didn't split anyway).

6537

Will change it to 'get a better occupancy'

[arsenm]

Can you add some tests stressing very wide VGPRs with calls? These will be dynamic vector indexing, and some image intrinsics (or you could just use inline asm)

llvm/docs/AMDGPUUsage.rst

6511

We have few contexts with single VPGR tuples > 4. Some image instructions and indirect register indexing. We probably don’t have any tests with calls stressing these cases

[cdevadas]

llvm/docs/AMDGPUUsage.rst

6511

Are you saying, there are cases more than 4 contiguous VGPRs to be allocated by RA? If yes, can you tell me the instructions?
I was under the impression that we have instructions requiring at most 4 contiguous VGPRs (for instance, FLAT_LOAD_DWORDX4)

[t-tye]

llvm/docs/AMDGPUUsage.rst

6511

No I am not talking about argument passing, I am talking about generating code for the body of the function. It would presumably be best that it only uses clobbered registers for values that do not want to be live across the calls it makes. So for objects larger than 4 dwords that will mean prologue/epilogue spilling/restoring, and register shuffling at call sites.

6511

We do have T# and V# values using in image instructions that are larger than 4 dwords. The instructions that use them currently need them in SGPRs but they may need to be moved from VGPRs (which would not need them to be contiguous). Do we also want to be sure the ABI will work for future hardware that may change, or are we ok with different ABIs for different targets?

[t-tye]

Added @mjbedy to review.

[rampitec]

llvm/docs/AMDGPUUsage.rst

6511

There are some image instructions which have address in much longer VGPR tuple. If you look into SiRegisterInfo.td our largest VGPR register class is 32 dwords long. I'd say this also shall be a minimal interleave value.

[arsenm]

llvm/docs/AMDGPUUsage.rst

6511

But those are AGPRs, which last I checked were not considered preserved across calls anyway. The question is also how often do those need to be live across calls, which is probably very rare.

[rampitec]

llvm/docs/AMDGPUUsage.rst

6511

AGPRs can be copied to VGPRs. Anyway, there are image addresses too which are pretty big.

[cdevadas]

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

[rampitec]

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

[arsenm]

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

[rampitec]

In D76356#1939646, @arsenm wrote:

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

OK. What if such a register needs to be preserved by a caller? I guess there is no safe window for it, so it will be forced to spill. Then we will spill a whole huge register, not just a part of it, because we do not use sublane spilling (except to AGPRs), right?

[arsenm]

In D76356#1939683, @rampitec wrote:

In D76356#1939646, @arsenm wrote:

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

OK. What if such a register needs to be preserved by a caller? I guess there is no safe window for it, so it will be forced to spill. Then we will spill a whole huge register, not just a part of it, because we do not use sublane spilling (except to AGPRs), right?

Yes, that's what I would expect

[rampitec]

In D76356#1942070, @arsenm wrote:

In D76356#1939683, @rampitec wrote:

In D76356#1939646, @arsenm wrote:

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

OK. What if such a register needs to be preserved by a caller? I guess there is no safe window for it, so it will be forced to spill. Then we will spill a whole huge register, not just a part of it, because we do not use sublane spilling (except to AGPRs), right?

Yes, that's what I would expect

Then probably interleave 4 is not a best choice. We may also need to adjust cost of tuples to make aligned allocation more likely.

[cdevadas]

In D76356#1942097, @rampitec wrote:

In D76356#1942070, @arsenm wrote:

In D76356#1939683, @rampitec wrote:

In D76356#1939646, @arsenm wrote:

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

OK. What if such a register needs to be preserved by a caller? I guess there is no safe window for it, so it will be forced to spill. Then we will spill a whole huge register, not just a part of it, because we do not use sublane spilling (except to AGPRs), right?

Yes, that's what I would expect

Then probably interleave 4 is not a best choice. We may also need to adjust cost of tuples to make aligned allocation more likely.

A cost for VGPR registers has been handled with https://reviews.llvm.org/D76417. This will ensure a balanced allocation of scratch registers & CSRs at every split (once the current patch is in the upstream).

[arsenm]

In D76356#1942097, @rampitec wrote:

In D76356#1942070, @arsenm wrote:

In D76356#1939683, @rampitec wrote:

In D76356#1939646, @arsenm wrote:

In D76356#1939580, @rampitec wrote:

In D76356#1939493, @cdevadas wrote:

Thank you all for the comments.
I can see that there are concerns with the current split boundary (4 VGPRs together), considering the fact that we have wide VGPR uses in certain scenarios (image instructions).
But, like Matt mentioned, how frequently such scenarios occur?
Changing the split boundary to a large value would probably take away the whole purpose of this patch - reduce the CSR spills & try to ensure a better occupancy.

What will happen if you need to pass VReg_1024 into a function? It might not be a common case, but will it even work?

It works, but that isn't changed by this patch. This is not changing the argument registers which are all still in v0-v31. I believe the largest argument type we pass in registers is <8 x i32>, and force <32 x i32> to be stack passed anyway

OK. What if such a register needs to be preserved by a caller? I guess there is no safe window for it, so it will be forced to spill. Then we will spill a whole huge register, not just a part of it, because we do not use sublane spilling (except to AGPRs), right?

Yes, that's what I would expect

Then probably interleave 4 is not a best choice. We may also need to adjust cost of tuples to make aligned allocation more likely.

It's a tradeoff between more CSR spills in general, and the case where a large tuple needs to live across a call (which I still expect to be extremely rare). AGPRs are never treated as CSRs anyway (and for the real AGPR use cases, calls are unlikely to be used)

[rampitec]

In D76356#1944135, @cdevadas wrote:

In D76356#1942097, @rampitec wrote:

Then probably interleave 4 is not a best choice. We may also need to adjust cost of tuples to make aligned allocation more likely.

A cost for VGPR registers has been handled with https://reviews.llvm.org/D76417. This will ensure a balanced allocation of scratch registers & CSRs at every split (once the current patch is in the upstream).

D76417 changes the cost of VGPRs, I am speaking about tuples. Assume you interleave by 4. Not you can allocate something into v[64:68] or v[65:69]. If you use the latter you will have a spill of 4 registers around any call site where it is alive. If you use the former you have a chance not to spill it.

[mjbedy]

llvm/docs/AMDGPUUsage.rst

6511

An additional concern I would have is with additional register fragmentation this might cause. This is essentially introducing an alignment requirement for wider VGPR allocations in these cases, which could cause higher register requirements. Has there been any investigation of the impact of this?

[cdevadas]

Divided the VGPRs into equal number of CSRs and scratch registers. Also, added a test case for VGPR tuple allocation.

[cdevadas]

Ping

[arsenm]

LGTM with test nits. This isn't perfect but we can't do much better now

llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll

7

inreg currently doesn't do anything for non-shaders, so you should remove it here.

9

Typo presreved

[cdevadas]

Fixed the testcase.

[t-tye]

LGTM. The call convention is still open to further refinement as more information is collected, but this appears to be an improvement so a good starting point.

[thakis]

This breaks check-llvm on Windows: http://45.33.8.238/win/14566/step_11.txt

Please take a look, and revert if it takes a while to investigate. (Maybe the test just needs a triple?)

llvm/docs/AMDGPUUsage.rst

 
 1.  VGPR0-31 and SGPR4-29 are used to pass function result arguments as
     described below. Any registers used are considered clobbered registers.
 2.  The following registers are preserved and have the same value as on entry:
 
     * FLAT_SCRATCH
     * EXEC
     * GFX6-8: M0
-    * All SGPR and VGPR registers except the clobbered registers of SGPR4-31 and
-      VGPR0-31.
+    * All SGPR registers except the clobbered registers of SGPR4-31.
+    * VGPR40-47
+      VGPR56-63
+      VGPR72-79
+      VGPR88-95
+      VGPR104-111
+      VGPR120-127
+      VGPR136-143
+      VGPR152-159
+      VGPR168-175
+      VGPR184-191
+      VGPR200-207
+      VGPR216-223
+      VGPR232-239
+      VGPR248-255
+        *Except the argument registers, the VGPR cloberred and the preserved
+        registers are intermixed at regular intervals in order to
+        get a better occupancy.*
 
       For the AMDGPU backend, an inter-procedural register allocation (IPRA)
-      optimization may mark some of clobbered SGPR4-31 and VGPR0-31 registers as
+      optimization may mark some of clobbered SGPR and VGPR registers as
       preserved if it can be determined that the called function does not change
       their value.
 
 2.  The PC is set to the RA provided on entry.
 3.  MODE register: *TBD*.
 4.  All other registers are clobbered.
 5.  Any necessary ``waitcnt`` has been performed to ensure memory accessed by
     function is available to the caller.
 
 .. TODO::
 
   - On gfx908 are all ACC registers clobbered?
 
   - How are function results returned? The address of structured types is passed
     by reference, but what about other types?

llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td

 def CSR_AMDGPU_VGPRs_24_255 : CalleeSavedRegs<
   (sequence "VGPR%u", 24, 255)
 >;
 
 def CSR_AMDGPU_VGPRs_32_255 : CalleeSavedRegs<
   (sequence "VGPR%u", 32, 255)
 >;
 
+def CSR_AMDGPU_VGPRs : CalleeSavedRegs<
+  // The CSRs & scratch-registers are interleaved at a split boundary of 8.
+  (add (sequence "VGPR%u", 40, 47),
+    (sequence "VGPR%u", 56, 63),
+    (sequence "VGPR%u", 72, 79),
+    (sequence "VGPR%u", 88, 95),
+    (sequence "VGPR%u", 104, 111),
+    (sequence "VGPR%u", 120, 127),
+    (sequence "VGPR%u", 136, 143),
+    (sequence "VGPR%u", 152, 159),
+    (sequence "VGPR%u", 168, 175),
+    (sequence "VGPR%u", 184, 191),
+    (sequence "VGPR%u", 200, 207),
+    (sequence "VGPR%u", 216, 223),
+    (sequence "VGPR%u", 232, 239),
+    (sequence "VGPR%u", 248, 255))
+>;
+
 def CSR_AMDGPU_SGPRs_32_105 : CalleeSavedRegs<
   (sequence "SGPR%u", 32, 105)
 >;
 
 // Just to get the regmask, not for calling convention purposes.
 def CSR_AMDGPU_AllVGPRs : CalleeSavedRegs<
   (sequence "VGPR%u", 0, 255)
 >;
 
 // Just to get the regmask, not for calling convention purposes.
 def CSR_AMDGPU_AllAllocatableSRegs : CalleeSavedRegs<
   (add (sequence "SGPR%u", 0, 105), VCC_LO, VCC_HI)
 >;
 
 def CSR_AMDGPU_HighRegs : CalleeSavedRegs<
-  (add CSR_AMDGPU_VGPRs_32_255, CSR_AMDGPU_SGPRs_32_105)
+  (add CSR_AMDGPU_VGPRs, CSR_AMDGPU_SGPRs_32_105)
 >;
 
 // Calling convention for leaf functions
 def CC_AMDGPU_Func : CallingConv<[
   CCIfByVal<CCPassByVal<4, 4>>,
   CCIfType<[i1], CCPromoteToType<i32>>,
   CCIfType<[i1, i8, i16], CCIfExtend<CCPromoteToType<i32>>>,
   CCIfType<[i32, f32, i16, f16, v2i16, v2f16, i1], CCAssignToReg<[

llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.ll

 }
 
 define void @dyn_insertelement_v8f64_const_s_v_v(double %val, i32 %idx) {
 ; GPRIDX-LABEL: dyn_insertelement_v8f64_const_s_v_v:
 ; GPRIDX:       ; %bb.0: ; %entry
 ; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GPRIDX-NEXT:    s_mov_b32 s18, 0
 ; GPRIDX-NEXT:    s_mov_b32 s19, 0x40200000
-; GPRIDX-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-; GPRIDX-NEXT:    buffer_store_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
-; GPRIDX-NEXT:    buffer_store_dword v34, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GPRIDX-NEXT:    s_mov_b32 s17, 0x401c0000
 ; GPRIDX-NEXT:    s_mov_b32 s16, s18
 ; GPRIDX-NEXT:    s_mov_b32 s15, 0x40180000
 ; GPRIDX-NEXT:    s_mov_b32 s14, s18
 ; GPRIDX-NEXT:    s_mov_b32 s13, 0x40140000
 ; GPRIDX-NEXT:    s_mov_b32 s12, s18
 ; GPRIDX-NEXT:    s_mov_b64 s[10:11], 4.0
 ; GPRIDX-NEXT:    s_mov_b32 s9, 0x40080000

 ; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
 ; GPRIDX-NEXT:    s_cbranch_execnz BB13_1
 ; GPRIDX-NEXT:  ; %bb.2:
 ; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
 ; GPRIDX-NEXT:    global_store_dwordx4 v[0:1], v[3:6], off
 ; GPRIDX-NEXT:    global_store_dwordx4 v[0:1], v[7:10], off
 ; GPRIDX-NEXT:    global_store_dwordx4 v[0:1], v[11:14], off
 ; GPRIDX-NEXT:    global_store_dwordx4 v[0:1], v[15:18], off
-; GPRIDX-NEXT:    buffer_load_dword v34, off, s[0:3], s32 ; 4-byte Folded Reload
-; GPRIDX-NEXT:    buffer_load_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
-; GPRIDX-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; GPRIDX-NEXT:    s_waitcnt vmcnt(0)
 ; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; MOVREL-LABEL: dyn_insertelement_v8f64_const_s_v_v:
 ; MOVREL:       ; %bb.0: ; %entry
 ; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; MOVREL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; MOVREL-NEXT:    s_mov_b32 s18, 0
 ; MOVREL-NEXT:    s_mov_b32 s19, 0x40200000
 ; MOVREL-NEXT:    s_mov_b32 s17, 0x401c0000
 ; MOVREL-NEXT:    s_mov_b32 s15, 0x40180000
 ; MOVREL-NEXT:    s_mov_b32 s13, 0x40140000
 ; MOVREL-NEXT:    s_mov_b32 s16, s18
 ; MOVREL-NEXT:    s_mov_b32 s14, s18
 ; MOVREL-NEXT:    s_mov_b32 s12, s18
 ; MOVREL-NEXT:    s_mov_b64 s[10:11], 4.0
 ; MOVREL-NEXT:    s_mov_b32 s9, 0x40080000
 ; MOVREL-NEXT:    s_mov_b32 s8, s18
 ; MOVREL-NEXT:    s_mov_b64 s[6:7], 2.0
 ; MOVREL-NEXT:    s_mov_b64 s[4:5], 1.0
-; MOVREL-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-; MOVREL-NEXT:    buffer_store_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
-; MOVREL-NEXT:    buffer_store_dword v34, off, s[0:3], s32 ; 4-byte Folded Spill
 ; MOVREL-NEXT:    v_mov_b32_e32 v34, s19
 ; MOVREL-NEXT:    v_mov_b32_e32 v33, s18
 ; MOVREL-NEXT:    v_mov_b32_e32 v32, s17
 ; MOVREL-NEXT:    v_mov_b32_e32 v31, s16
 ; MOVREL-NEXT:    v_mov_b32_e32 v30, s15
 ; MOVREL-NEXT:    v_mov_b32_e32 v29, s14
 ; MOVREL-NEXT:    v_mov_b32_e32 v28, s13
 ; MOVREL-NEXT:    v_mov_b32_e32 v27, s12

 ; MOVREL-NEXT:    s_xor_b32 exec_lo, exec_lo, vcc_lo
 ; MOVREL-NEXT:    s_cbranch_execnz BB13_1
 ; MOVREL-NEXT:  ; %bb.2:
 ; MOVREL-NEXT:    s_mov_b32 exec_lo, s4
 ; MOVREL-NEXT:    global_store_dwordx4 v[0:1], v[3:6], off
 ; MOVREL-NEXT:    global_store_dwordx4 v[0:1], v[7:10], off
 ; MOVREL-NEXT:    global_store_dwordx4 v[0:1], v[11:14], off
 ; MOVREL-NEXT:    global_store_dwordx4 v[0:1], v[15:18], off
-; MOVREL-NEXT:    buffer_load_dword v34, off, s[0:3], s32 ; 4-byte Folded Reload
-; MOVREL-NEXT:    buffer_load_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
-; MOVREL-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
-; MOVREL-NEXT:    s_waitcnt vmcnt(0)
 ; MOVREL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; MOVREL-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %insert = insertelement <8 x double> <double 1.0, double 2.0, double 3.0, double 4.0, double 5.0, double 6.0, double 7.0, double 8.0>, double %val, i32 %idx
   %vec.0 = shufflevector <8 x double> %insert, <8 x double> undef, <2 x i32> <i32 0, i32 1>
   %vec.1 = shufflevector <8 x double> %insert, <8 x double> undef, <2 x i32> <i32 2, i32 3>
   %vec.2 = shufflevector <8 x double> %insert, <8 x double> undef, <2 x i32> <i32 4, i32 5>
   %vec.3 = shufflevector <8 x double> %insert, <8 x double> undef, <2 x i32> <i32 6, i32 7>

llvm/test/CodeGen/AMDGPU/call-argument-types.ll

 define amdgpu_kernel void @stack_passed_arg_alignment_v32i32_f64(<32 x i32> %val, double %tmp) #0 {
 entry:
   call void @stack_passed_f64_arg(<32 x i32> %val, double %tmp)
   ret void
 }
 
 ; GCN-LABEL: {{^}}tail_call_byval_align16:
 ; GCN-NOT: s32
-; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
-; GCN: buffer_store_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:20
-; GCN: buffer_load_dword v33, off, s[0:3], s32 offset:16
+; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:12
+; GCN: buffer_load_dword v33, off, s[0:3], s32 offset:8
 
 ; GCN: s_getpc_b64
 
 ; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:4
 ; GCN: buffer_store_dword v33, off, s[0:3], s32{{$}}
-; GCN: buffer_load_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
-; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
 ; GCN-NOT: s32
 ; GCN: s_setpc_b64
 define void @tail_call_byval_align16(<32 x i32> %val, double %tmp) #0 {
 entry:
   %alloca = alloca double, align 8, addrspace(5)
   tail call void @byval_align16_f64_arg(<32 x i32> %val, double addrspace(5)* byval align 16 %alloca)
   ret void
 }
 
 ; GCN-LABEL: {{^}}tail_call_stack_passed_arg_alignment_v32i32_f64:
 ; GCN-NOT: s32
-; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
-; GCN: buffer_store_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
 ; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:4
 ; GCN: buffer_load_dword v33, off, s[0:3], s32{{$}}
 ; GCN: s_getpc_b64
 ; GCN: buffer_store_dword v33, off, s[0:3], s32{{$}}
 ; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:4
-; GCN: buffer_load_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
-; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
 ; GCN-NOT: s32
 ; GCN: s_setpc_b64
 define void @tail_call_stack_passed_arg_alignment_v32i32_f64(<32 x i32> %val, double %tmp) #0 {
 entry:
   tail call void @stack_passed_f64_arg(<32 x i32> %val, double %tmp)
   ret void
 }
 

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

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mattr=-code-object-v3 -enable-ipra=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CI %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mattr=-code-object-v3 -mcpu=fiji -enable-ipra=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,VI-NOBUG %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mattr=-code-object-v3 -mcpu=iceland -enable-ipra=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,VI-BUG %s
 
 ; Make sure to run a GPU with the SGPR allocation bug.
 
 ; GCN-LABEL: {{^}}use_vcc:
 ; GCN: ; NumSgprs: 34
 ; GCN: ; NumVgprs: 0
 define void @use_vcc() #1 {
   call void asm sideeffect "", "~{vcc}" () #0
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_use_vcc:
-; GCN: v_writelane_b32 v32, s33, 2
-; GCN: v_writelane_b32 v32, s30, 0
-; GCN: v_writelane_b32 v32, s31, 1
+; GCN: v_writelane_b32 v40, s33, 2
+; GCN: v_writelane_b32 v40, s30, 0
+; GCN: v_writelane_b32 v40, s31, 1
 ; GCN: s_swappc_b64
-; GCN: v_readlane_b32 s4, v32, 0
-; GCN: v_readlane_b32 s5, v32, 1
-; GCN: v_readlane_b32 s33, v32, 2
+; GCN: v_readlane_b32 s4, v40, 0
+; GCN: v_readlane_b32 s5, v40, 1
+; GCN: v_readlane_b32 s33, v40, 2
 ; GCN: ; NumSgprs: 36
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define void @indirect_use_vcc() #1 {
   call void @use_vcc()
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_2level_use_vcc_kernel:
 ; GCN: is_dynamic_callstack = 0
 ; CI: ; NumSgprs: 38
 ; VI-NOBUG: ; NumSgprs: 40
 ; VI-BUG: ; NumSgprs: 96
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define amdgpu_kernel void @indirect_2level_use_vcc_kernel(i32 addrspace(1)* %out) #0 {
   call void @indirect_use_vcc()
   ret void
 }
 
 ; GCN-LABEL: {{^}}use_flat_scratch:
 ; CI: ; NumSgprs: 36
 ; VI: ; NumSgprs: 38
 ; GCN: ; NumVgprs: 0
 define void @use_flat_scratch() #1 {
   call void asm sideeffect "", "~{flat_scratch}" () #0
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_use_flat_scratch:
 ; CI: ; NumSgprs: 38
 ; VI: ; NumSgprs: 40
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define void @indirect_use_flat_scratch() #1 {
   call void @use_flat_scratch()
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_2level_use_flat_scratch_kernel:
 ; GCN: is_dynamic_callstack = 0
 ; CI: ; NumSgprs: 38
 ; VI-NOBUG: ; NumSgprs: 40
 ; VI-BUG: ; NumSgprs: 96
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define amdgpu_kernel void @indirect_2level_use_flat_scratch_kernel(i32 addrspace(1)* %out) #0 {
   call void @indirect_use_flat_scratch()
   ret void
 }
 
 ; GCN-LABEL: {{^}}use_10_vgpr:
 ; GCN: ; NumVgprs: 10
 define void @use_10_vgpr() #1 {
   call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4}"() #0
   call void asm sideeffect "", "~{v5},~{v6},~{v7},~{v8},~{v9}"() #0
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_use_10_vgpr:
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define void @indirect_use_10_vgpr() #0 {
   call void @use_10_vgpr()
   ret void
 }
 
 ; GCN-LABEL: {{^}}indirect_2_level_use_10_vgpr:
 ; GCN: is_dynamic_callstack = 0
-; GCN: ; NumVgprs: 33
+; GCN: ; NumVgprs: 41
 define amdgpu_kernel void @indirect_2_level_use_10_vgpr() #0 {
   call void @indirect_use_10_vgpr()
   ret void
 }
 
-; GCN-LABEL: {{^}}use_40_vgpr:
-; GCN: ; NumVgprs: 40
-define void @use_40_vgpr() #1 {
-  call void asm sideeffect "", "~{v39}"() #0
+; GCN-LABEL: {{^}}use_50_vgpr:
+; GCN: ; NumVgprs: 50
+define void @use_50_vgpr() #1 {
+  call void asm sideeffect "", "~{v49}"() #0
   ret void
 }
 
-; GCN-LABEL: {{^}}indirect_use_40_vgpr:
-; GCN: ; NumVgprs: 40
-define void @indirect_use_40_vgpr() #0 {
-  call void @use_40_vgpr()
+; GCN-LABEL: {{^}}indirect_use_50_vgpr:
+; GCN: ; NumVgprs: 50
+define void @indirect_use_50_vgpr() #0 {
+  call void @use_50_vgpr()
   ret void
 }
 
 ; GCN-LABEL: {{^}}use_80_sgpr:
 ; GCN: ; NumSgprs: 80
 define void @use_80_sgpr() #1 {
   call void asm sideeffect "", "~{s79}"() #0
   ret void

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

define amdgpu_kernel void @test_kernel_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void() #0 {
   call void @external_void_func_void()
   call void asm sideeffect "", ""() #0
   call void @external_void_func_void()
   ret void
 }
 
 ; 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, 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
+; GCN: v_writelane_b32 v40, s33, 4
+; GCN: v_writelane_b32 v40, s34, 0
+; GCN: v_writelane_b32 v40, s35, 1
+; GCN: v_writelane_b32 v40, s30, 2
+; GCN: v_writelane_b32 v40, s31, 3
 
 ; GCN: s_swappc_b64
 ; GCN-NEXT: ;;#ASMSTART
 ; GCN-NEXT: ;;#ASMEND
 ; 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 s35, v32, 1
-; GCN: v_readlane_b32 s34, v32, 0
+; GCN-DAG: v_readlane_b32 s4, v40, 2
+; GCN-DAG: v_readlane_b32 s5, v40, 3
+; GCN: v_readlane_b32 s35, v40, 1
+; GCN: v_readlane_b32 s34, v40, 0
 
-; GCN: v_readlane_b32 s33, v32, 4
+; GCN: v_readlane_b32 s33, v40, 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 {
   call void @external_void_func_void()
   call void asm sideeffect "", ""() #0
   call void @external_void_func_void()
   ret void
 }
 
 ; GCN-LABEL: {{^}}test_func_call_external_void_funcx2:
-; GCN: buffer_store_dword v32
-; GCN: v_writelane_b32 v32, s33, 4
+; GCN: buffer_store_dword v40
+; GCN: v_writelane_b32 v40, s33, 4
 
 ; 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 s33, v32, 4
-; GCN: buffer_load_dword v32,
+; GCN: v_readlane_b32 s33, v40, 4
+; GCN: buffer_load_dword v40,
 define void @test_func_call_external_void_funcx2() #0 {
   call void @external_void_func_void()
   call void @external_void_func_void()
   ret void
 }
 
 ; GCN-LABEL: {{^}}void_func_void_clobber_s30_s31:
 ; GCN: s_waitcnt

 define amdgpu_kernel void @test_call_void_func_void_mayclobber_s31(i32 addrspace(1)* %out) #0 {
   %s31 = call i32 asm sideeffect "; def $0", "={s31}"()
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "{s31}"(i32 %s31)
   ret void
 }
 
 ; GCN-LABEL: {{^}}test_call_void_func_void_mayclobber_v31:
-; GCN: v_mov_b32_e32 v32, v31
+; GCN: v_mov_b32_e32 v40, v31
 ; GCN-NEXT: s_swappc_b64
-; GCN-NEXT: v_mov_b32_e32 v31, v32
+; GCN-NEXT: v_mov_b32_e32 v31, v40
 define amdgpu_kernel void @test_call_void_func_void_mayclobber_v31(i32 addrspace(1)* %out) #0 {
   %v31 = call i32 asm sideeffect "; def $0", "={v31}"()
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "{v31}"(i32 %v31)
   ret void
 }
 
 ; FIXME: What is the expected behavior for reserved registers here?

 ; GCN-NEXT: s_endpgm
 define amdgpu_kernel void @test_call_void_func_void_preserves_s34(i32 addrspace(1)* %out) #0 {
   %s34 = call i32 asm sideeffect "; def $0", "={s34}"()
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "{s34}"(i32 %s34)
   ret void
 }
 
-; GCN-LABEL: {{^}}test_call_void_func_void_preserves_v32: {{.*}}
+; GCN-LABEL: {{^}}test_call_void_func_void_preserves_v40: {{.*}}
 
 ; GCN-NOT: v32
 ; GCN: s_getpc_b64 s[4:5]
 ; GCN-NEXT: s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
 ; GCN-NEXT: s_addc_u32 s5, s5, external_void_func_void@rel32@hi+4
 ; GCN: s_mov_b32 s32, 0
-; GCN-NOT: v32
+; GCN-NOT: v40
 
 ; GCN: ;;#ASMSTART
-; GCN-NEXT: ; def v32
+; GCN-NEXT: ; def v40
 ; GCN-NEXT: ;;#ASMEND
 
 ; GCN: s_swappc_b64 s[30:31], s[4:5]
 
-; GCN-NOT: v32
+; GCN-NOT: v40
 
 ; GCN: ;;#ASMSTART
-; GCN-NEXT: ; use v32
+; GCN-NEXT: ; use v40
 ; GCN-NEXT: ;;#ASMEND
 ; GCN-NEXT: s_endpgm
-define amdgpu_kernel void @test_call_void_func_void_preserves_v32(i32 addrspace(1)* %out) #0 {
-  %v32 = call i32 asm sideeffect "; def $0", "={v32}"()
+define amdgpu_kernel void @test_call_void_func_void_preserves_v40(i32 addrspace(1)* %out) #0 {
+  %v40 = call i32 asm sideeffect "; def $0", "={v40}"()
   call void @external_void_func_void()
-  call void asm sideeffect "; use $0", "{v32}"(i32 %v32)
+  call void asm sideeffect "; use $0", "{v40}"(i32 %v40)
   ret void
 }
 
 ; GCN-LABEL: {{^}}void_func_void_clobber_s33:
 ; GCN: v_writelane_b32 v0, s33, 0
 ; GCN-NEXT: #ASMSTART
 ; GCN-NEXT: ; clobber
 ; GCN-NEXT: #ASMEND

 ; GCN-NEXT: s_endpgm
 define amdgpu_kernel void @test_call_void_func_void_clobber_s34() #0 {
   call void @void_func_void_clobber_s34()
   ret void
 }
 
 ; GCN-LABEL: {{^}}callee_saved_sgpr_func:
 ; GCN-NOT: s40
-; GCN: v_writelane_b32 v32, s40
+; GCN: v_writelane_b32 v40, s40
 ; GCN: s_swappc_b64
 ; GCN-NOT: s40
 ; GCN: ; use s40
 ; GCN-NOT: s40
-; GCN: v_readlane_b32 s40, v32
+; GCN: v_readlane_b32 s40, v40
 ; GCN-NOT: s40
 define void @callee_saved_sgpr_func() #2 {
   %s40 = call i32 asm sideeffect "; def s40", "={s40}"() #0
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "s"(i32 %s40) #0
   ret void
 }
 
define amdgpu_kernel void @callee_saved_sgpr_kernel() #2 {
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "s"(i32 %s40) #0
   ret void
 }
 
 ; First call preserved VGPR is used so it can't be used for SGPR spills.
 ; GCN-LABEL: {{^}}callee_saved_sgpr_vgpr_func:
 ; GCN-NOT: s40
-; GCN: v_writelane_b32 v33, s40
+; GCN: v_writelane_b32 v41, s40
 ; GCN: s_swappc_b64
 ; GCN-NOT: s40
 ; GCN: ; use s40
 ; GCN-NOT: s40
-; GCN: v_readlane_b32 s40, v33
+; GCN: v_readlane_b32 s40, v41
 ; GCN-NOT: s40
 define void @callee_saved_sgpr_vgpr_func() #2 {
   %s40 = call i32 asm sideeffect "; def s40", "={s40}"() #0
-  %v32 = call i32 asm sideeffect "; def v32", "={v32}"() #0
+  %v40 = call i32 asm sideeffect "; def v40", "={v40}"() #0
   call void @external_void_func_void()
   call void asm sideeffect "; use $0", "s"(i32 %s40) #0
-  call void asm sideeffect "; use $0", "v"(i32 %v32) #0
+  call void asm sideeffect "; use $0", "v"(i32 %v40) #0
   ret void
 }
 
 ; GCN-LABEL: {{^}}callee_saved_sgpr_vgpr_kernel:
 ; GCN-NOT: s40
 ; GCN: ; def s40
 ; GCN-NOT: s40
 ; GCN: s_swappc_b64

llvm/test/CodeGen/AMDGPU/call-waitcnt.ll

 ; GCN-NEXT:    s_addc_u32 flat_scratch_hi, s7, 0
 ; GCN-NEXT:    s_add_u32 s0, s0, s9
 ; GCN-NEXT:    s_addc_u32 s1, s1, 0
 ; GCN-NEXT:    v_mov_b32_e32 v0, 0
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, func@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, func@rel32@hi+4
 ; GCN-NEXT:    s_mov_b32 s32, 0
-; GCN-NEXT:    v_mov_b32_e32 v32, 0
+; GCN-NEXT:    v_mov_b32_e32 v40, 0
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GCN-NEXT:    v_mov_b32_e32 v0, s34
 ; GCN-NEXT:    v_mov_b32_e32 v1, s35
-; GCN-NEXT:    global_store_dword v[0:1], v32, off
+; GCN-NEXT:    global_store_dword v[0:1], v40, off
 ; GCN-NEXT:    s_endpgm
   call void @func(i32 0)
   store i32 0, i32 addrspace(1)* %ptr
   ret void
 }
 
 define amdgpu_kernel void @call_no_wait_after_call_return_val(i32 addrspace(1)* %ptr, i32) #0 {
 ; GCN-LABEL: call_no_wait_after_call_return_val:

llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll

 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
 ; GCN-DAG: s_add_u32 s32, s32, 0x400
 ; GCN-DAG: v_writelane_b32 [[CSR_VGPR]], s33, [[FP_SPILL_LANE:[0-9]+]]
 
 ; GCN-DAG: v_writelane_b32 [[CSR_VGPR]], s30, 0
 ; GCN-DAG: v_writelane_b32 [[CSR_VGPR]], s31, 1
 ; GCN: s_swappc_b64
 
-; GCN-DAG: v_readlane_b32 s4, v32, 0
-; GCN-DAG: v_readlane_b32 s5, v32, 1
+; GCN-DAG: v_readlane_b32 s4, v40, 0
+; GCN-DAG: v_readlane_b32 s5, v40, 1
 
 ; GCN: s_sub_u32 s32, s32, 0x400
 ; GCN-NEXT: v_readlane_b32 s33, [[CSR_VGPR]], [[FP_SPILL_LANE]]
 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GCN-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64

 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt
 ; GCN-NEXT: s_setpc_b64
 define void @callee_func_sgpr_spill_no_calls(i32 %in) #0 {
   call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
   call void asm sideeffect "", "~{v8},~{v9},~{v10},~{v11},~{v12},~{v13},~{v14},~{v15}"() #0
   call void asm sideeffect "", "~{v16},~{v17},~{v18},~{v19},~{v20},~{v21},~{v22},~{v23}"() #0
   call void asm sideeffect "", "~{v24},~{v25},~{v26},~{v27},~{v28},~{v29},~{v30},~{v31}"() #0
+  call void asm sideeffect "", "~{v32},~{v33},~{v34},~{v35},~{v36},~{v37},~{v38},~{v39}"() #0
 
   %wide.sgpr0 = call <16 x i32> asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr1 = call <16 x i32> asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr2 = call <16 x i32> asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr5 = call <16 x i32> asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr3 = call <8 x i32> asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr4 = call <2 x i32> asm sideeffect "; def $0", "=s" () #0
 

 }
 
 ; TODO: Can the SP inc/deec be remvoed?
 ; GCN-LABEL: {{^}}callee_with_stack_no_fp_elim_csr_vgpr:
 ; GCN: s_waitcnt
 ; GCN-NEXT:s_mov_b32 [[FP_COPY:s[0-9]+]], s33
 ; GCN-NEXT: s_mov_b32 s33, s32
 ; GCN: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0
-; GCN-DAG: buffer_store_dword v33, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-DAG: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-DAG: buffer_store_dword [[ZERO]], off, s[0:3], s33 offset:8
 
 ; GCN:  ;;#ASMSTART
-; GCN-NEXT: ; clobber v33
+; GCN-NEXT: ; clobber v41
 ; GCN-NEXT: ;;#ASMEND
 
-; GCN: buffer_load_dword v33, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN: s_add_u32 s32, s32, 0x300
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x300
 ; GCN-NEXT: s_mov_b32 s33, s4
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @callee_with_stack_no_fp_elim_csr_vgpr() #1 {
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, i32 addrspace(5)* %alloca
-  call void asm sideeffect "; clobber v33", "~{v33}"()
+  call void asm sideeffect "; clobber v41", "~{v41}"()
   ret void
 }
 
 ; Use a copy to a free SGPR instead of introducing a second CSR VGPR.
 ; GCN-LABEL: {{^}}last_lane_vgpr_for_fp_csr:
 ; GCN: s_waitcnt
 ; GCN-NEXT: v_writelane_b32 v1, s33, 63
 ; GCN-NEXT: s_mov_b32 s33, s32
-; GCN: buffer_store_dword v33, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-COUNT-63: v_writelane_b32 v1
 ; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s33 offset:8
 ; GCN: ;;#ASMSTART
 ; GCN-COUNT-63: v_readlane_b32 s{{[0-9]+}}, v1
 
 ; GCN: s_add_u32 s32, s32, 0x300
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x300
 ; GCN-NEXT: v_readlane_b32 s33, v1, 63
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @last_lane_vgpr_for_fp_csr() #1 {
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, i32 addrspace(5)* %alloca
-  call void asm sideeffect "; clobber v33", "~{v33}"()
+  call void asm sideeffect "; clobber v41", "~{v41}"()
   call void asm sideeffect "",
     "~{s40},~{s41},~{s42},~{s43},~{s44},~{s45},~{s46},~{s47},~{s48},~{s49}
     ,~{s50},~{s51},~{s52},~{s53},~{s54},~{s55},~{s56},~{s57},~{s58},~{s59}
     ,~{s60},~{s61},~{s62},~{s63},~{s64},~{s65},~{s66},~{s67},~{s68},~{s69}
     ,~{s70},~{s71},~{s72},~{s73},~{s74},~{s75},~{s76},~{s77},~{s78},~{s79}
     ,~{s80},~{s81},~{s82},~{s83},~{s84},~{s85},~{s86},~{s87},~{s88},~{s89}
     ,~{s90},~{s91},~{s92},~{s93},~{s94},~{s95},~{s96},~{s97},~{s98},~{s99}
     ,~{s100},~{s101},~{s102}"() #1
 
   ret void
 }
 
 ; Use a copy to a free SGPR instead of introducing a second CSR VGPR.
 ; GCN-LABEL: {{^}}no_new_vgpr_for_fp_csr:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_mov_b32 [[FP_COPY:s[0-9]+]], s33
 ; GCN-NEXT: s_mov_b32 s33, s32
-; GCN-NEXT: buffer_store_dword v33, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-COUNT-64: v_writelane_b32 v1,
 
 ; GCN: buffer_store_dword
 ; GCN: ;;#ASMSTART
 ; GCN-COUNT-64: v_readlane_b32 s{{[0-9]+}}, v1
 
-; GCN: buffer_load_dword v33, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN: s_add_u32 s32, s32, 0x300
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x300
 ; GCN-NEXT: s_mov_b32 s33, [[FP_COPY]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @no_new_vgpr_for_fp_csr() #1 {
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, i32 addrspace(5)* %alloca
-  call void asm sideeffect "; clobber v33", "~{v33}"()
+  call void asm sideeffect "; clobber v41", "~{v41}"()
   call void asm sideeffect "",
     "~{s39},~{s40},~{s41},~{s42},~{s43},~{s44},~{s45},~{s46},~{s47},~{s48},~{s49}
     ,~{s50},~{s51},~{s52},~{s53},~{s54},~{s55},~{s56},~{s57},~{s58},~{s59}
     ,~{s60},~{s61},~{s62},~{s63},~{s64},~{s65},~{s66},~{s67},~{s68},~{s69}
     ,~{s70},~{s71},~{s72},~{s73},~{s74},~{s75},~{s76},~{s77},~{s78},~{s79}
     ,~{s80},~{s81},~{s82},~{s83},~{s84},~{s85},~{s86},~{s87},~{s88},~{s89}
     ,~{s90},~{s91},~{s92},~{s93},~{s94},~{s95},~{s96},~{s97},~{s98},~{s99}
     ,~{s100},~{s101},~{s102}"() #1

 }
 
 ; Need a new CSR VGPR to satisfy the FP spill.
 ; GCN-LABEL: {{^}}no_unused_non_csr_sgpr_for_fp_no_scratch_vgpr:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GCN-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
-; GCN-NEXT: v_writelane_b32 v32, s33, 2
-; GCN-NEXT: v_writelane_b32 v32, s30, 0
+; GCN-NEXT: v_writelane_b32 [[CSR_VGPR]], s33, 2
+; GCN-NEXT: v_writelane_b32 [[CSR_VGPR]], s30, 0
 ; GCN-NEXT: s_mov_b32 s33, s32
 
-; GCN-DAG: v_writelane_b32 v32, s31, 1
+; GCN-DAG: v_writelane_b32 [[CSR_VGPR]], s31, 1
 ; GCN-DAG: buffer_store_dword
 ; GCN: s_add_u32 s32, s32, 0x300{{$}}
 
 ; GCN: ;;#ASMSTART
 
-; GCN: v_readlane_b32 s4, v32, 0
-; GCN-NEXT: v_readlane_b32 s5, v32, 1
+; GCN: v_readlane_b32 s4, [[CSR_VGPR]], 0
+; GCN-NEXT: v_readlane_b32 s5, [[CSR_VGPR]], 1
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x300{{$}}
-; GCN-NEXT: v_readlane_b32 s33, v32, 2
+; GCN-NEXT: v_readlane_b32 s33, [[CSR_VGPR]], 2
 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GCN-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @no_unused_non_csr_sgpr_for_fp_no_scratch_vgpr() #1 {
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, i32 addrspace(5)* %alloca
 
   ; Use all clobberable registers, so FP has to spill to a VGPR.
   call void asm sideeffect "",
     "~{s0},~{s1},~{s2},~{s3},~{s4},~{s5},~{s6},~{s7},~{s8},~{s9}
     ,~{s10},~{s11},~{s12},~{s13},~{s14},~{s15},~{s16},~{s17},~{s18},~{s19}
     ,~{s20},~{s21},~{s22},~{s23},~{s24},~{s25},~{s26},~{s27},~{s28},~{s29}
     ,~{s30},~{s31}"() #0
 
-  call void asm sideeffect "; clobber nonpreserved VGPRs",
+  call void asm sideeffect "; clobber nonpreserved initial VGPRs",
     "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
     ,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
     ,~{v20},~{v21},~{v22},~{v23},~{v24},~{v25},~{v26},~{v27},~{v28},~{v29}
-    ,~{v30},~{v31}"() #1
+    ,~{v30},~{v31},~{v32},~{v33},~{v34},~{v35},~{v36},~{v37},~{v38},~{v39}"() #1
 
   ret void
 }
 
 ; The byval argument exceeds the MUBUF constant offset, so a scratch
 ; register is needed to access the CSR VGPR slot.
 ; GCN-LABEL: {{^}}scratch_reg_needed_mubuf_offset:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GCN-NEXT: v_mov_b32_e32 [[SCRATCH_VGPR:v[0-9]+]], 0x1008
 ; GCN-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], [[SCRATCH_VGPR]], s[0:3], s32 offen ; 4-byte Folded Spill
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
-; GCN-NEXT: v_writelane_b32 v32, s33, 2
-; GCN-NEXT: v_writelane_b32 v32, s30, 0
+; GCN-NEXT: v_writelane_b32 [[CSR_VGPR]], s33, 2
+; GCN-NEXT: v_writelane_b32 [[CSR_VGPR]], s30, 0
 ; GCN-NEXT: s_mov_b32 s33, s32
-; GCN-DAG: v_writelane_b32 v32, s31, 1
+; GCN-DAG: v_writelane_b32 [[CSR_VGPR]], s31, 1
 ; GCN-DAG: s_add_u32 s32, s32, 0x40300{{$}}
 ; GCN-DAG: buffer_store_dword
 
 ; GCN: ;;#ASMSTART
 
-; GCN: v_readlane_b32 s4, v32, 0
-; GCN-NEXT: v_readlane_b32 s5, v32, 1
+; GCN: v_readlane_b32 s4, [[CSR_VGPR]], 0
+; GCN-NEXT: v_readlane_b32 s5, [[CSR_VGPR]], 1
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x40300{{$}}
-; GCN-NEXT: v_readlane_b32 s33, v32, 2
+; GCN-NEXT: v_readlane_b32 s33, [[CSR_VGPR]], 2
 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GCN-NEXT: v_mov_b32_e32 [[SCRATCH_VGPR:v[0-9]+]], 0x1008
 ; GCN-NEXT: buffer_load_dword [[CSR_VGPR]], [[SCRATCH_VGPR]], s[0:3], s32 offen ; 4-byte Folded Reload
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @scratch_reg_needed_mubuf_offset([4096 x i8] addrspace(5)* byval align 4 %arg) #1 {
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, i32 addrspace(5)* %alloca
 
   ; Use all clobberable registers, so FP has to spill to a VGPR.
   call void asm sideeffect "; clobber nonpreserved SGPRs",
     "~{s0},~{s1},~{s2},~{s3},~{s4},~{s5},~{s6},~{s7},~{s8},~{s9}
     ,~{s10},~{s11},~{s12},~{s13},~{s14},~{s15},~{s16},~{s17},~{s18},~{s19}
     ,~{s20},~{s21},~{s22},~{s23},~{s24},~{s25},~{s26},~{s27},~{s28},~{s29}
     ,~{s30},~{s31}"() #0
 
   ; Use all clobberable VGPRs, so a CSR spill is needed for the VGPR
   call void asm sideeffect "; clobber nonpreserved VGPRs",
     "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
     ,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
     ,~{v20},~{v21},~{v22},~{v23},~{v24},~{v25},~{v26},~{v27},~{v28},~{v29}
-    ,~{v30},~{v31}"() #1
+    ,~{v30},~{v31},~{v32},~{v33},~{v34},~{v35},~{v36},~{v37},~{v38},~{v39}"() #1
 
   ret void
 }
 
 ; GCN-LABEL: {{^}}local_empty_func:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_setpc_b64
 define internal void @local_empty_func() #0 {

llvm/test/CodeGen/AMDGPU/callee-special-input-sgprs-fixed-abi.ll

define hidden void @use_workgroup_id_yz() #1 {
   ret void
 }
 
 ; Argument is in right place already
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_x:
 ; GCN-NOT: s12
 ; GCN-NOT: s13
 ; GCN-NOT: s14
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_x() #1 {
   call void @use_workgroup_id_x()
   ret void
 }
 
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_y:
 ; GCN-NOT: s4
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_y() #1 {
   call void @use_workgroup_id_y()
   ret void
 }
 
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_z:
 ; GCN-NOT: s4
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_z() #1 {
   call void @use_workgroup_id_z()
   ret void
 }
 
 ; GCN-LABEL: {{^}}other_arg_use_workgroup_id_x:
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
 ; GCN: ; use s12

llvm/test/CodeGen/AMDGPU/callee-special-input-sgprs.ll

 define amdgpu_kernel void @kern_indirect_use_workgroup_id_yz() #1 {
   call void @use_workgroup_id_yz()
   ret void
 }
 
 ; Argument is in right place already
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_x:
 ; GCN-NOT: s4
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_x() #1 {
   call void @use_workgroup_id_x()
   ret void
 }
 
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_y:
 ; GCN-NOT: s4
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_y() #1 {
   call void @use_workgroup_id_y()
   ret void
 }
 
 ; GCN-LABEL: {{^}}func_indirect_use_workgroup_id_z:
 ; GCN-NOT: s4
-; GCN: v_readlane_b32 s4, v32, 0
+; GCN: v_readlane_b32 s4, v40, 0
 define hidden void @func_indirect_use_workgroup_id_z() #1 {
   call void @use_workgroup_id_z()
   ret void
 }
 
 ; GCN-LABEL: {{^}}other_arg_use_workgroup_id_x:
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
 ; GCN: ; use s4

llvm/test/CodeGen/AMDGPU/callee-special-input-vgprs.ll

 ; FIXEDABI: v_or_b32_e32 [[TMP2:v[0-9]+]], v0, [[TMP1]]
 ; FIXEDABI: v_or_b32_e32 v31, [[TMP2]], [[TMP0]]
 define amdgpu_kernel void @kern_indirect_other_arg_use_workitem_id_z() #1 {
   call void @other_arg_use_workitem_id_z(i32 555)
   ret void
 }
 
 ; GCN-LABEL: {{^}}too_many_args_use_workitem_id_x:
-; VARABI: buffer_store_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
 ; VARABI: buffer_load_dword v32, off, s[0:3], s32{{$}}
 ; VARABI: v_and_b32_e32 v32, 0x3ff, v32
 ; VARABI: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, v32
 
-; VARABI: buffer_load_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
-; VARABI-NEXT: s_waitcnt
+; VARABI: s_waitcnt
 ; VARABI-NEXT: s_setpc_b64
 
 ; FIXEDABI: v_and_b32_e32 v31, 0x3ff, v31
 ; FIXEDABI: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s32{{$}}
 define void @too_many_args_use_workitem_id_x(
   i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7,
   i32 %arg8, i32 %arg9, i32 %arg10, i32 %arg11, i32 %arg12, i32 %arg13, i32 %arg14, i32 %arg15,
   i32 %arg16, i32 %arg17, i32 %arg18, i32 %arg19, i32 %arg20, i32 %arg21, i32 %arg22, i32 %arg23,
call void @too_many_args_use_workitem_id_x(
     i32 250, i32 260, i32 270, i32 280,
     i32 290, i32 300, i32 310, i32 320)
   ret void
 }
 
 ; Requires loading and storing to stack slot.
 ; GCN-LABEL: {{^}}too_many_args_call_too_many_args_use_workitem_id_x:
 ; GCN-DAG: s_add_u32 s32, s32, 0x400{{$}}
-; GCN-DAG: buffer_store_dword v32, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GCN-DAG: buffer_store_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
 ; GCN-DAG: buffer_load_dword v32, off, s[0:3], s33{{$}}
 
 ; GCN: buffer_store_dword v32, off, s[0:3], s32{{$}}
 
 ; GCN: s_swappc_b64
 
-; GCN: buffer_load_dword v32, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 ; GCN: s_sub_u32 s32, s32, 0x400{{$}}
+; GCN: buffer_load_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
 ; GCN: s_setpc_b64
 define void @too_many_args_call_too_many_args_use_workitem_id_x(
   i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7,
   i32 %arg8, i32 %arg9, i32 %arg10, i32 %arg11, i32 %arg12, i32 %arg13, i32 %arg14, i32 %arg15,
   i32 %arg16, i32 %arg17, i32 %arg18, i32 %arg19, i32 %arg20, i32 %arg21, i32 %arg22, i32 %arg23,
   i32 %arg24, i32 %arg25, i32 %arg26, i32 %arg27, i32 %arg28, i32 %arg29, i32 %arg30, i32 %arg31) #1 {
   call void @too_many_args_use_workitem_id_x(
     i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7,
     i32 %arg8, i32 %arg9, i32 %arg10, i32 %arg11, i32 %arg12, i32 %arg13, i32 %arg14, i32 %arg15,
     i32 %arg16, i32 %arg17, i32 %arg18, i32 %arg19, i32 %arg20, i32 %arg21, i32 %arg22, i32 %arg23,
     i32 %arg24, i32 %arg25, i32 %arg26, i32 %arg27, i32 %arg28, i32 %arg29, i32 %arg30, i32 %arg31)
   ret void
 }
 
 ; var abi stack layout:
 ; frame[0] = byval arg32
 ; frame[1] = stack passed workitem ID x
 ; frame[2] = VGPR spill slot
 
 ; GCN-LABEL: {{^}}too_many_args_use_workitem_id_x_byval:
-; VARABI: buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
 ; VARABI: buffer_load_dword v32, off, s[0:3], s32 offset:4
 ; VARABI-NEXT: s_waitcnt
 ; VARABI-NEXT: v_and_b32_e32 v32, 0x3ff, v32
 ; VARABI-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, v32
 ; VARABI: buffer_load_dword v0, off, s[0:3], s32{{$}}
-; VARABI: buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; VARABI: s_setpc_b64
 
 
 ; FIXEDABI: v_and_b32_e32 v31, 0x3ff, v31
 ; FIXEDABI-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, v31
 
 ; FIXEDABI: buffer_load_dword v0, off, s[0:3], s32{{$}}
 ; FIXEDABI: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
call void @too_many_args_use_workitem_id_x_byval(
     i32 170, i32 180, i32 190, i32 200,
     i32 210, i32 220, i32 230, i32 240,
     i32 250, i32 260, i32 270, i32 280,
     i32 290, i32 300, i32 310, i32 320,
     i32 addrspace(5)* %alloca)
   ret void
 }
 
-; Only one stack load should be emitted for all 3 values.
 ; GCN-LABEL: {{^}}too_many_args_use_workitem_id_xyz:
-; VARABI: buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-; VARABI: buffer_store_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
 ; VARABI-NOT: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s32{{$}}
 ; VARABI: buffer_load_dword v32, off, s[0:3], s32{{$}}
 ; VARABI-NOT: buffer_load_dword
 
 ; VARABI: v_and_b32_e32 [[AND_X:v[0-9]+]], 0x3ff, v32
 ; VARABI-NOT: buffer_load_dword
 ; VARABI: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[AND_X]]
 ; VARABI-NOT: buffer_load_dword
 ; VARABI: v_bfe_u32 [[BFE_Y:v[0-9]+]], v32, 10, 10
 ; VARABI-NEXT: v_bfe_u32 [[BFE_Z:v[0-9]+]], v32, 20, 10
 ; VARABI-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[BFE_Y]]
 ; VARABI-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[BFE_Z]]
 
-; VARABI: buffer_load_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
-; VARABI: buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
-; VARABI-NEXT: s_waitcnt
+; VARABI: s_waitcnt
 ; VARABI-NEXT: s_setpc_b64
 
 
 ; FIXEDABI: v_and_b32_e32 [[AND_X:v[0-9]+]], 0x3ff, v31
 ; FIXEDABI-NOT: buffer_load_dword
 ; FIXEDABI: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[AND_X]]
 ; FIXEDABI-NOT: buffer_load_dword
 ; FIXEDABI: v_bfe_u32 [[BFE_Y:v[0-9]+]], v31, 10, 10

 ; GCN-DAG: {{flat|global}}_store_dword v[0:1], [[IDX]]
 ; GCN-DAG: v_bfe_u32 [[IDY:v[0-9]+]], v31, 10, 10
 ; GCN-DAG: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[IDY]]
 ; GCN-DAG: v_bfe_u32 [[IDZ:v[0-9]+]], v31, 20, 10
 ; GCN-DAG: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[IDZ]]
 
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_setpc_b64
-; GCN: ScratchSize: 8
+; GCN: ScratchSize: 0
 define void @too_many_args_use_workitem_id_x_stack_yz(
   i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7,
   i32 %arg8, i32 %arg9, i32 %arg10, i32 %arg11, i32 %arg12, i32 %arg13, i32 %arg14, i32 %arg15,
   i32 %arg16, i32 %arg17, i32 %arg18, i32 %arg19, i32 %arg20, i32 %arg21, i32 %arg22, i32 %arg23,
   i32 %arg24, i32 %arg25, i32 %arg26, i32 %arg27, i32 %arg28, i32 %arg29, i32 %arg30) #1 {
   %val0 = call i32 @llvm.amdgcn.workitem.id.x()
   store volatile i32 %val0, i32 addrspace(1)* undef
   %val1 = call i32 @llvm.amdgcn.workitem.id.y()

llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll

 ; since that didn't look like an ABI copy.
 
 
 define float @call_split_type_used_outside_block_v2f32() #0 {
 ; GCN-LABEL: call_split_type_used_outside_block_v2f32:
 ; GCN:       ; %bb.0: ; %bb0
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GCN-NEXT:    buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
-; GCN-NEXT:    v_writelane_b32 v32, s33, 2
-; GCN-NEXT:    v_writelane_b32 v32, s30, 0
+; GCN-NEXT:    v_writelane_b32 v40, s33, 2
+; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_add_u32 s32, s32, 0x400
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, func_v2f32@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, func_v2f32@rel32@hi+4
-; GCN-NEXT:    v_writelane_b32 v32, s31, 1
+; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; GCN-NEXT:    v_readlane_b32 s4, v32, 0
-; GCN-NEXT:    v_readlane_b32 s5, v32, 1
+; GCN-NEXT:    v_readlane_b32 s4, v40, 0
+; GCN-NEXT:    v_readlane_b32 s5, v40, 1
 ; GCN-NEXT:    s_sub_u32 s32, s32, 0x400
-; GCN-NEXT:    v_readlane_b32 s33, v32, 2
+; GCN-NEXT:    v_readlane_b32 s33, v40, 2
 ; GCN-NEXT:    s_or_saveexec_b64 s[6:7], -1
-; GCN-NEXT:    buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[4:5]
 bb0:
   %split.ret.type = call <2 x float> @func_v2f32()
   br label %bb1
 
 bb1:
   %extract = extractelement <2 x float> %split.ret.type, i32 0
   ret float %extract
 }
 
 define float @call_split_type_used_outside_block_v3f32() #0 {
 ; GCN-LABEL: call_split_type_used_outside_block_v3f32:
 ; GCN:       ; %bb.0: ; %bb0
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GCN-NEXT:    buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
-; GCN-NEXT:    v_writelane_b32 v32, s33, 2
-; GCN-NEXT:    v_writelane_b32 v32, s30, 0
+; GCN-NEXT:    v_writelane_b32 v40, s33, 2
+; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_add_u32 s32, s32, 0x400
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, func_v3f32@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, func_v3f32@rel32@hi+4
-; GCN-NEXT:    v_writelane_b32 v32, s31, 1
+; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; GCN-NEXT:    v_readlane_b32 s4, v32, 0
-; GCN-NEXT:    v_readlane_b32 s5, v32, 1
+; GCN-NEXT:    v_readlane_b32 s4, v40, 0
+; GCN-NEXT:    v_readlane_b32 s5, v40, 1
 ; GCN-NEXT:    s_sub_u32 s32, s32, 0x400
-; GCN-NEXT:    v_readlane_b32 s33, v32, 2
+; GCN-NEXT:    v_readlane_b32 s33, v40, 2
 ; GCN-NEXT:    s_or_saveexec_b64 s[6:7], -1
-; GCN-NEXT:    buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[4:5]
 bb0:
   %split.ret.type = call <3 x float> @func_v3f32()
   br label %bb1
 
 bb1:
   %extract = extractelement <3 x float> %split.ret.type, i32 0
   ret float %extract
 }
 
 define half @call_split_type_used_outside_block_v4f16() #0 {
 ; GCN-LABEL: call_split_type_used_outside_block_v4f16:
 ; GCN:       ; %bb.0: ; %bb0
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GCN-NEXT:    buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
-; GCN-NEXT:    v_writelane_b32 v32, s33, 2
-; GCN-NEXT:    v_writelane_b32 v32, s30, 0
+; GCN-NEXT:    v_writelane_b32 v40, s33, 2
+; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_add_u32 s32, s32, 0x400
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, func_v4f16@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, func_v4f16@rel32@hi+4
-; GCN-NEXT:    v_writelane_b32 v32, s31, 1
+; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; GCN-NEXT:    v_readlane_b32 s4, v32, 0
-; GCN-NEXT:    v_readlane_b32 s5, v32, 1
+; GCN-NEXT:    v_readlane_b32 s4, v40, 0
+; GCN-NEXT:    v_readlane_b32 s5, v40, 1
 ; GCN-NEXT:    s_sub_u32 s32, s32, 0x400
-; GCN-NEXT:    v_readlane_b32 s33, v32, 2
+; GCN-NEXT:    v_readlane_b32 s33, v40, 2
 ; GCN-NEXT:    s_or_saveexec_b64 s[6:7], -1
-; GCN-NEXT:    buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[4:5]
 bb0:
   %split.ret.type = call <4 x half> @func_v4f16()
   br label %bb1
 
 bb1:
   %extract = extractelement <4 x half> %split.ret.type, i32 0
   ret half %extract
 }
 
 define { i32, half } @call_split_type_used_outside_block_struct() #0 {
 ; GCN-LABEL: call_split_type_used_outside_block_struct:
 ; GCN:       ; %bb.0: ; %bb0
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GCN-NEXT:    buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
-; GCN-NEXT:    v_writelane_b32 v32, s33, 2
-; GCN-NEXT:    v_writelane_b32 v32, s30, 0
+; GCN-NEXT:    v_writelane_b32 v40, s33, 2
+; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_add_u32 s32, s32, 0x400
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, func_struct@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, func_struct@rel32@hi+4
-; GCN-NEXT:    v_writelane_b32 v32, s31, 1
+; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; GCN-NEXT:    v_readlane_b32 s4, v32, 0
-; GCN-NEXT:    v_readlane_b32 s5, v32, 1
+; GCN-NEXT:    v_readlane_b32 s4, v40, 0
+; GCN-NEXT:    v_readlane_b32 s5, v40, 1
 ; GCN-NEXT:    v_mov_b32_e32 v1, v4
 ; GCN-NEXT:    s_sub_u32 s32, s32, 0x400
-; GCN-NEXT:    v_readlane_b32 s33, v32, 2
+; GCN-NEXT:    v_readlane_b32 s33, v40, 2
 ; GCN-NEXT:    s_or_saveexec_b64 s[6:7], -1
-; GCN-NEXT:    buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[4:5]
 bb0:
   %split.ret.type = call { <4 x i32>, <4 x half> } @func_struct()
   br label %bb1
 
 bb1:

llvm/test/CodeGen/AMDGPU/ipra-regmask.ll

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -enable-ipra -print-regusage -o /dev/null 2>&1 < %s | FileCheck %s
 ; Make sure the expected regmask is generated for sub/superregisters.
 
 ; CHECK-DAG: csr Clobbered Registers: $vgpr0 $vgpr0_hi16 $vgpr0_lo16 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr0_vgpr1 $vgpr0_vgpr1_vgpr2 {{$}}
 define void @csr() #0 {
-  call void asm sideeffect "", "~{v0},~{v36},~{v37}"() #0
+  call void asm sideeffect "", "~{v0},~{v44},~{v45}"() #0
   ret void
 }
 
 ; CHECK-DAG: subregs_for_super Clobbered Registers: $vgpr0 $vgpr1 $vgpr0_hi16 $vgpr1_hi16 $vgpr0_lo16 $vgpr1_lo16 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16 $vgpr0_vgpr1 $vgpr1_vgpr2 $vgpr0_vgpr1_vgpr2 $vgpr1_vgpr2_vgpr3 {{$}}
 define void @subregs_for_super() #0 {
   call void asm sideeffect "", "~{v0},~{v1}"() #0
   ret void
 }

llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll

; CHECK-NOT: mul i32
   ret void
 }
 
 define void @slsr1_1(i32 %b.arg, i32 %s.arg) #0 {
 ; GFX9-LABEL: slsr1_1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX9-NEXT:    buffer_store_dword v35, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX9-NEXT:    buffer_store_dword v43, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX9-NEXT:    v_writelane_b32 v35, s33, 4
+; GFX9-NEXT:    v_writelane_b32 v43, s33, 4
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_add_u32 s32, s32, 0x800
-; GFX9-NEXT:    buffer_store_dword v32, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
-; GFX9-NEXT:    buffer_store_dword v33, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
-; GFX9-NEXT:    buffer_store_dword v34, off, s[0:3], s33 ; 4-byte Folded Spill
-; GFX9-NEXT:    v_writelane_b32 v35, s34, 0
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX9-NEXT:    buffer_store_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT:    buffer_store_dword v42, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT:    v_writelane_b32 v43, s34, 0
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, foo@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, foo@gotpcrel32@hi+4
-; GFX9-NEXT:    v_writelane_b32 v35, s35, 1
+; GFX9-NEXT:    v_writelane_b32 v43, s35, 1
 ; GFX9-NEXT:    s_load_dwordx2 s[34:35], s[4:5], 0x0
-; GFX9-NEXT:    v_mov_b32_e32 v32, v1
-; GFX9-NEXT:    v_mov_b32_e32 v33, v0
-; GFX9-NEXT:    v_writelane_b32 v35, s30, 2
-; GFX9-NEXT:    v_mul_u32_u24_e32 v0, v33, v32
-; GFX9-NEXT:    v_writelane_b32 v35, s31, 3
-; GFX9-NEXT:    v_and_b32_e32 v34, 0xffffff, v32
+; GFX9-NEXT:    v_mov_b32_e32 v40, v1
+; GFX9-NEXT:    v_mov_b32_e32 v41, v0
+; GFX9-NEXT:    v_writelane_b32 v43, s30, 2
+; GFX9-NEXT:    v_mul_u32_u24_e32 v0, v41, v40
+; GFX9-NEXT:    v_writelane_b32 v43, s31, 3
+; GFX9-NEXT:    v_and_b32_e32 v42, 0xffffff, v40
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_swappc_b64 s[30:31], s[34:35]
-; GFX9-NEXT:    v_mad_u32_u24 v32, v33, v32, v34
-; GFX9-NEXT:    v_mov_b32_e32 v0, v32
+; GFX9-NEXT:    v_mad_u32_u24 v40, v41, v40, v42
+; GFX9-NEXT:    v_mov_b32_e32 v0, v40
 ; GFX9-NEXT:    s_swappc_b64 s[30:31], s[34:35]
-; GFX9-NEXT:    v_add_u32_e32 v0, v32, v34
+; GFX9-NEXT:    v_add_u32_e32 v0, v40, v42
 ; GFX9-NEXT:    s_swappc_b64 s[30:31], s[34:35]
-; GFX9-NEXT:    v_readlane_b32 s4, v35, 2
-; GFX9-NEXT:    v_readlane_b32 s5, v35, 3
-; GFX9-NEXT:    v_readlane_b32 s35, v35, 1
-; GFX9-NEXT:    v_readlane_b32 s34, v35, 0
-; GFX9-NEXT:    buffer_load_dword v34, off, s[0:3], s33 ; 4-byte Folded Reload
-; GFX9-NEXT:    buffer_load_dword v33, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
-; GFX9-NEXT:    buffer_load_dword v32, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GFX9-NEXT:    v_readlane_b32 s4, v43, 2
+; GFX9-NEXT:    v_readlane_b32 s5, v43, 3
+; GFX9-NEXT:    v_readlane_b32 s35, v43, 1
+; GFX9-NEXT:    v_readlane_b32 s34, v43, 0
+; GFX9-NEXT:    buffer_load_dword v42, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT:    buffer_load_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_sub_u32 s32, s32, 0x800
-; GFX9-NEXT:    v_readlane_b32 s33, v35, 4
+; GFX9-NEXT:    v_readlane_b32 s33, v43, 4
 ; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
-; GFX9-NEXT:    buffer_load_dword v35, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GFX9-NEXT:    buffer_load_dword v43, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[4:5]
   %b = and i32 %b.arg, 16777215
   %s = and i32 %s.arg, 16777215
 
 ; CHECK-LABEL: @slsr1(
   ; foo(b * s);

llvm/test/CodeGen/AMDGPU/nested-calls.ll

 ; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -march=amdgcn -mcpu=hawaii -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=CI %s
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX9 -check-prefix=VI %s
 
 ; Test calls when called by other callable functions rather than
 ; kernels.
 
 declare void @external_void_func_i32(i32) #0
 
 ; GCN-LABEL: {{^}}test_func_call_external_void_func_i32_imm:
 ; GCN: s_waitcnt
 
 ; Spill CSR VGPR used for SGPR spilling
 ; GCN: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
-; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
-; GCN-DAG: v_writelane_b32 v32, s33, 2
+; GCN-DAG: v_writelane_b32 v40, s33, 2
 ; GCN-DAG: s_mov_b32 s33, s32
 ; GCN-DAG: s_add_u32 s32, s32, 0x400
-; GCN-DAG: v_writelane_b32 v32, s30, 0
-; GCN-DAG: v_writelane_b32 v32, s31, 1
+; GCN-DAG: v_writelane_b32 v40, s30, 0
+; GCN-DAG: v_writelane_b32 v40, s31, 1
 
 ; GCN: s_swappc_b64
 
-; GCN: v_readlane_b32 s4, v32, 0
-; GCN: v_readlane_b32 s5, v32, 1
+; GCN: v_readlane_b32 s4, v40, 0
+; GCN: v_readlane_b32 s5, v40, 1
 
 ; GCN-NEXT: s_sub_u32 s32, s32, 0x400
-; GCN-NEXT: v_readlane_b32 s33, v32, 2
+; GCN-NEXT: v_readlane_b32 s33, v40, 2
 ; GCN: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
-; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64 s[4:5]
 define void @test_func_call_external_void_func_i32_imm() #0 {
   call void @external_void_func_i32(i32 42)
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/regbank-reassign.mir

bb.0:
     GLOBAL_STORE_DWORDX4 %3, %6, 0, 0, 0, 0, implicit $exec
     GLOBAL_STORE_DWORDX4 %3, %7, 0, 0, 0, 0, implicit $exec
     GLOBAL_STORE_DWORDX4 %3, %8, 0, 0, 0, 0, implicit $exec
     GLOBAL_STORE_DWORDX4 %3, %9, 0, 0, 0, 0, implicit $exec
     S_ENDPGM 0
 ...
 
 # GCN-LABEL: csr{{$}}
-# GCN: V_AND_B32_e32 $vgpr4, $vgpr0,
+# GCN: V_AND_B32_e32 $vgpr37, $vgpr0,
 ---
 name:            csr
 tracksRegLiveness: true
 registers:
   - { id: 0, class: vgpr_32, preferred-register: '$vgpr0' }
   - { id: 1, class: vgpr_32, preferred-register: '$vgpr4' }
   - { id: 2, class: vgpr_32, preferred-register: '$vgpr1' }
   - { id: 3, class: vreg_64, preferred-register: '$vgpr2_vgpr3' }

llvm/test/CodeGen/AMDGPU/sibling-call.ll

define fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %arg0, i32 %arg1, [32 x i32] %large) #1 {
   %add1 = add i32 %add0, %val_firststack
   %add2 = add i32 %add1, %val_laststack
   ret i32 %add2
 }
 
 ; FIXME: Why load and store same location for stack args?
 ; GCN-LABEL: {{^}}sibling_call_i32_fastcc_i32_i32_a32i32:
 
-; GCN-DAG: buffer_store_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
-; GCN-DAG: buffer_store_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-
 ; GCN-DAG: buffer_load_dword [[LOAD_0:v[0-9]+]], off, s[0:3], s32{{$}}
 ; GCN-DAG: buffer_load_dword [[LOAD_1:v[0-9]+]], off, s[0:3], s32 offset:4
 
 ; GCN-NOT: s32
 
 ; GCN-DAG: buffer_store_dword [[LOAD_0]], off, s[0:3], s32{{$}}
 ; GCN-DAG: buffer_store_dword [[LOAD_1]], off, s[0:3], s32 offset:4
 
-; GCN-DAG: buffer_load_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
-; GCN-DAG: buffer_load_dword v33, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
-
 ; GCN-NOT: s32
 ; GCN: s_setpc_b64
 define fastcc i32 @sibling_call_i32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] %c) #1 {
 entry:
   %ret = tail call fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] %c)
   ret i32 %ret
 }
 
 ; GCN-LABEL: {{^}}sibling_call_i32_fastcc_i32_i32_a32i32_stack_object:
 ; GCN-DAG: v_mov_b32_e32 [[NINE:v[0-9]+]], 9
-; GCN: buffer_store_dword [[NINE]], off, s[0:3], s32 offset:40
+; GCN: buffer_store_dword [[NINE]], off, s[0:3], s32 offset:28
 ; GCN: s_setpc_b64
 define fastcc i32 @sibling_call_i32_fastcc_i32_i32_a32i32_stack_object(i32 %a, i32 %b, [32 x i32] %c) #1 {
 entry:
   %alloca = alloca [16 x i32], align 4, addrspace(5)
   %gep = getelementptr inbounds [16 x i32], [16 x i32] addrspace(5)* %alloca, i32 0, i32 5
   store volatile i32 9, i32 addrspace(5)* %gep
   %ret = tail call fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] %c)
   ret i32 %ret

 entry:
   %ret = tail call fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] zeroinitializer)
   ret i32 %ret
 }
 
 ; Have another non-tail in the function
 ; GCN-LABEL: {{^}}sibling_call_i32_fastcc_i32_i32_other_call:
 ; GCN: s_or_saveexec_b64 s{{\[[0-9]+:[0-9]+\]}}, -1
-; GCN-NEXT: buffer_store_dword v34, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
 ; GCN-NEXT: s_mov_b64 exec
 ; GCN: s_mov_b32 s33, s32
 ; GCN-DAG: s_add_u32 s32, s32, 0x400
 
-; GCN-DAG: buffer_store_dword v32, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
-; GCN-DAG: buffer_store_dword v33, off, s[0:3], s33 ; 4-byte Folded Spill
-; GCN-DAG: v_writelane_b32 v34, s34, 0
-; GCN-DAG: v_writelane_b32 v34, s35, 1
+; GCN-DAG: buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GCN-DAG: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-DAG: v_writelane_b32 v42, s34, 0
+; GCN-DAG: v_writelane_b32 v42, s35, 1
 
 ; GCN-DAG: s_getpc_b64 s[4:5]
 ; GCN-DAG: s_add_u32 s4, s4, i32_fastcc_i32_i32@gotpcrel32@lo+4
 ; GCN-DAG: s_addc_u32 s5, s5, i32_fastcc_i32_i32@gotpcrel32@hi+4
 
 
 ; GCN: s_swappc_b64
 
-; GCN-DAG: v_readlane_b32 s34, v34, 0
-; GCN-DAG: v_readlane_b32 s35, v34, 1
+; GCN-DAG: v_readlane_b32 s34, v42, 0
+; GCN-DAG: v_readlane_b32 s35, v42, 1
 
-; GCN: buffer_load_dword v33, off, s[0:3], s33 ; 4-byte Folded Reload
-; GCN: buffer_load_dword v32, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GCN: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN: buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 
 ; GCN: s_getpc_b64 s[4:5]
 ; GCN-NEXT: s_add_u32 s4, s4, sibling_call_i32_fastcc_i32_i32@rel32@lo+4
 ; GCN-NEXT: s_addc_u32 s5, s5, sibling_call_i32_fastcc_i32_i32@rel32@hi+4
 
 ; GCN: s_sub_u32 s32, s32, 0x400
 ; GCN-NEXT: v_readlane_b32 s33,
 ; GCN-NEXT: s_or_saveexec_b64 s[6:7], -1
-; GCN-NEXT: buffer_load_dword v34, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; GCN-NEXT: s_mov_b64 exec, s[6:7]
 ; GCN-NEXT: s_setpc_b64 s[4:5]
 define fastcc i32 @sibling_call_i32_fastcc_i32_i32_other_call(i32 %a, i32 %b, i32 %c) #1 {
 entry:
   %other.call = tail call fastcc i32 @i32_fastcc_i32_i32(i32 %a, i32 %b)
   %ret = tail call fastcc i32 @sibling_call_i32_fastcc_i32_i32(i32 %a, i32 %b, i32 %other.call)
   ret i32 %ret
 }
 
 ; Have stack object in caller and stack passed arguments. SP should be
 ; in same place at function exit.
 
 ; GCN-LABEL: {{^}}sibling_call_stack_objecti32_fastcc_i32_i32_a32i32:
 ; GCN-NOT: s33
-; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32 offset:
+; GCN: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s32 offset:
 
 ; GCN-NOT: s33
 
-; GCN: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s32 offset:
+; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32 offset:
 ; GCN: s_setpc_b64 s[4:5]
 define fastcc i32 @sibling_call_stack_objecti32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] %c) #1 {
 entry:
   %alloca = alloca [16 x i32], align 4, addrspace(5)
   %gep = getelementptr inbounds [16 x i32], [16 x i32] addrspace(5)* %alloca, i32 0, i32 5
   store volatile i32 9, i32 addrspace(5)* %gep
   %ret = tail call fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %a, i32 %b, [32 x i32] %c)
   ret i32 %ret

llvm/test/CodeGen/AMDGPU/spill-csr-frame-ptr-reg-copy.ll

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -verify-machineinstrs -stress-regalloc=1 < %s | FileCheck -check-prefix=GCN %s
 
 ; GCN-LABEL: {{^}}spill_csr_s5_copy:
 ; GCN: s_or_saveexec_b64
-; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
 ; GCN-NEXT: s_mov_b64 exec
-; GCN: v_writelane_b32 v32, s33, 2
+; GCN: v_writelane_b32 v40, s33, 2
 ; GCN: s_swappc_b64
 
 ; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 9
 ; GCN: buffer_store_dword [[K]], off, s[0:3], s33{{$}}
 
-; GCN: v_readlane_b32 s33, v32, 2
+; GCN: v_readlane_b32 s33, v40, 2
 ; GCN: s_or_saveexec_b64
-; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
 ; GCN: s_mov_b64 exec
 ; GCN: s_setpc_b64
 define void @spill_csr_s5_copy() #0 {
 bb:
   %alloca = alloca i32, addrspace(5)
   %tmp = tail call i64 @func() #1
   %tmp1 = getelementptr inbounds i32, i32 addrspace(1)* null, i64 %tmp
   %tmp2 = load i32, i32 addrspace(1)* %tmp1, align 4

llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll

 ; GCN-NEXT:    s_addc_u32 s37, s37, 0
 ; GCN-NEXT:    v_mov_b32_e32 v1, 0x2000
 ; GCN-NEXT:    v_mov_b32_e32 v2, 0x4000
 ; GCN-NEXT:    v_mov_b32_e32 v3, 0
 ; GCN-NEXT:    v_mov_b32_e32 v4, 0x400000
 ; GCN-NEXT:    s_mov_b64 s[0:1], s[36:37]
 ; GCN-NEXT:    s_mov_b64 s[2:3], s[38:39]
 ; GCN-NEXT:    s_mov_b32 s32, 0xc0000
-; GCN-NEXT:    v_add_nc_u32_e64 v32, 4, 0x4000
+; GCN-NEXT:    v_add_nc_u32_e64 v40, 4, 0x4000
 ; GCN-NEXT:    ; implicit-def: $vcc_hi
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, svm_eval_nodes@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, svm_eval_nodes@rel32@hi+4
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_mov_b32_e32 v0, s6
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v0
 ; GCN-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GCN-NEXT:    s_cbranch_execz BB0_2
 ; GCN-NEXT:  ; %bb.1: ; %if.then4.i
-; GCN-NEXT:    buffer_load_dword v0, v32, s[36:39], s32 offen
-; GCN-NEXT:    buffer_load_dword v1, v32, s[36:39], s32 offen offset:4
+; GCN-NEXT:    buffer_load_dword v0, v40, s[36:39], s32 offen
+; GCN-NEXT:    buffer_load_dword v1, v40, s[36:39], s32 offen offset:4
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_add_nc_u32_e32 v0, v1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, 0x41c64e6d, v0
 ; GCN-NEXT:    v_add_nc_u32_e32 v0, 0x3039, v0
 ; GCN-NEXT:    buffer_store_dword v0, v0, s[36:39], 0 offen
 ; GCN-NEXT:  BB0_2: ; %shader_eval_surface.exit
 ; GCN-NEXT:    s_endpgm
 entry:

llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll

+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
+
+declare void @extern_func()
+
+define <4 x float> @non_preserved_vgpr_tuple8(<8 x i32> %rsrc, <4 x i32> %samp, float %bias, float %zcompare, float %s, float %t, float %clamp) {
+; The vgpr tuple8 operand in image_gather4_c_b_cl instruction needs not be
+; preserved across the call and should get 8 scratch registers.
+
+; GFX9-LABEL: non_preserved_vgpr_tuple8:
+; GFX9: buffer_store_dword v44, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9: buffer_store_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill
+
+; GFX9: v_mov_b32_e32 v37, v11
+; GFX9-NEXT: v_mov_b32_e32 v38, v10
+; GFX9-NEXT: v_mov_b32_e32 v49, v9
+; GFX9-NEXT: v_writelane_b32 v44, s30, 0
+; GFX9-NEXT: v_mov_b32_e32 v36, v16
+; GFX9-NEXT: v_mov_b32_e32 v35, v15
+; GFX9-NEXT: v_mov_b32_e32 v34, v14
+; GFX9-NEXT: v_mov_b32_e32 v33, v13
+; GFX9-NEXT: v_mov_b32_e32 v32, v12
+
+; GFX9: ;;#ASMSTART
+; GFX9-NEXT: ;;#ASMEND
+
+; GFX9: image_gather4_c_b_cl v[40:43], v[32:39], s[4:11], s[12:15] dmask:0x1
+; GFX9-NEXT: s_getpc_b64 s[4:5]
+; GFX9-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+4
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+
+; GFX9: buffer_load_dword v43, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GFX9: buffer_load_dword v44, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX9: s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: non_preserved_vgpr_tuple8:
+; GFX10: buffer_store_dword v44, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX10: buffer_store_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill
+
+; GFX10: v_mov_b32_e32 v36, v16
+; GFX10-NEXT: v_mov_b32_e32 v35, v15
+; GFX10-NEXT: v_mov_b32_e32 v34, v14
+; GFX10-NEXT: v_mov_b32_e32 v33, v13
+; GFX10-NEXT: v_mov_b32_e32 v32, v12
+
+; GFX10: ;;#ASMSTART
+; GFX10-NEXT: ;;#ASMEND
+
+; GFX10: image_gather4_c_b_cl v[40:43], v[32:39], s[4:11], s[12:15] dmask:0x1
+; GFX10-NEXT: v_nop
+; GFX10-NEXT: s_getpc_b64 s[4:5]
+; GFX10-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4
+; GFX10-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+4
+; GFX10-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5]
+
+; GFX10: buffer_load_dword v43, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+
+; GFX10: buffer_load_dword v44, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX10: s_setpc_b64 s[4:5]
+main_body:
+  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
+  call void asm sideeffect "", "~{v8},~{v9},~{v10},~{v11},~{v12},~{v13},~{v14},~{v15}"() #0
+  call void asm sideeffect "", "~{v16},~{v17},~{v18},~{v19},~{v20},~{v21},~{v22},~{v23}"() #0
+  call void asm sideeffect "", "~{v24},~{v25},~{v26},~{v27},~{v28},~{v29},~{v30},~{v31}"() #0
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f32(i32 1, float %bias, float %zcompare, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  call void @extern_func()
+  ret <4 x float> %v
+}
+
+define <4 x float> @call_preserved_vgpr_tuple8(<8 x i32> %rsrc, <4 x i32> %samp, float %bias, float %zcompare, float %s, float %t, float %clamp) {
+; The vgpr tuple8 operand in image_gather4_c_b_cl instruction needs to be preserved
+; across the call and should get allcoated to 8 CSRs.
+; Only the lower 5 sub-registers of the tuple are preserved.
+; The upper 3 sub-registers are unused.
+
+; GFX9-LABEL: call_preserved_vgpr_tuple8:
+; GFX9: buffer_store_dword v56, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GFX9: buffer_store_dword v40, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill
+
+; GFX9: v_mov_b32_e32 v44, v16
+; GFX9-NEXT: v_mov_b32_e32 v43, v15
+; GFX9-NEXT: v_mov_b32_e32 v42, v14
+; GFX9-NEXT: v_mov_b32_e32 v41, v13
+; GFX9-NEXT: v_mov_b32_e32 v40, v12
+
+; GFX9: s_getpc_b64 s[4:5]
+; GFX9-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+4
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT: image_gather4_c_b_cl v[0:3], v[40:47], s[36:43], s[44:47] dmask:0x1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_store_dwordx4 v[0:1], v[0:3], off
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: image_gather4_c_b_cl v[0:3], v[40:47], s[36:43], s[44:47] dmask:0x1
+
+; GFX9: buffer_load_dword v44, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+
+; GFX9: buffer_load_dword v56, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GFX9: s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: call_preserved_vgpr_tuple8:
+; GFX10: buffer_store_dword v45, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GFX10: buffer_store_dword v40, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill
+
+; GFX10: s_getpc_b64 s[4:5]
+; GFX10-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4
+; GFX10-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+4
+; GFX10-NEXT: v_mov_b32_e32 v40, v16
+; GFX10-NEXT: v_mov_b32_e32 v41, v15
+; GFX10-NEXT: image_gather4_c_b_cl v[0:3], v[12:19], s[36:43], s[44:47] dmask:0x1
+; GFX10-NEXT: v_mov_b32_e32 v42, v14
+; GFX10-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT: v_mov_b32_e32 v43, v13
+; GFX10-NEXT: v_mov_b32_e32 v44, v12
+; GFX10-NEXT: ; implicit-def: $vcc_hi
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: global_store_dwordx4 v[0:1], v[0:3], off
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT: image_gather4_c_b_cl v[0:3], [v44, v43, v42, v41, v40], s[36:43], s[44:47] dmask:0x1
+
+; GFX10: buffer_load_dword v44, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GFX10-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+; GFX10: buffer_load_dword v45, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GFX10: s_setpc_b64 s[4:5]
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f32(i32 1, float %bias, float %zcompare, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  store <4 x float> %v, <4 x float> addrspace(1)* undef
+  call void @extern_func()
+  %v1 = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f32(i32 1, float %bias, float %zcompare, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  ret <4 x float> %v1
+}
+
+declare <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f32(i32 immarg, float, float, float, float, float, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #1
+
+attributes #0 = { nounwind writeonly }
+attributes #1 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/virtregrewrite-undef-identity-copy.mir

 machineFunctionInfo:
   isEntryFunction: true
   scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
   frameOffsetReg:  '$sgpr95'
   stackPtrOffsetReg: '$sgpr32'
 body:             |
   bb.0:
     ; CHECK-LABEL: name: undef_identity_copy
-    ; CHECK: renamable $vgpr32_vgpr33_vgpr34_vgpr35 = FLAT_LOAD_DWORDX4 undef renamable $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16, addrspace 1)
+    ; CHECK: renamable $vgpr40_vgpr41_vgpr42_vgpr43 = FLAT_LOAD_DWORDX4 undef renamable $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16, addrspace 1)
     ; CHECK: renamable $sgpr6_sgpr7 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) @foo + 4, target-flags(amdgpu-rel32-hi) @foo + 4, implicit-def dead $scc
     ; CHECK: ADJCALLSTACKUP 0, 0, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr95
     ; CHECK: $sgpr4 = COPY $sgpr95
     ; CHECK: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr6_sgpr7, @foo, csr_amdgpu_highregs, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4
     ; CHECK: ADJCALLSTACKDOWN 0, 4, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr95
     ; CHECK: renamable $sgpr6_sgpr7 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) @bar + 4, target-flags(amdgpu-rel32-hi) @bar + 4, implicit-def dead $scc
     ; CHECK: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr95
     ; CHECK: $sgpr4 = COPY $sgpr95
-    ; CHECK: $vgpr0 = COPY renamable $vgpr32
-    ; CHECK: $vgpr1 = COPY renamable $vgpr33
-    ; CHECK: $vgpr2 = COPY renamable $vgpr34
+    ; CHECK: $vgpr0 = COPY renamable $vgpr40
+    ; CHECK: $vgpr1 = COPY renamable $vgpr41
+    ; CHECK: $vgpr2 = COPY renamable $vgpr42
     ; CHECK: $vgpr3 = KILL undef renamable $vgpr3
     ; CHECK: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr6_sgpr7, @bar, csr_amdgpu_highregs, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4, implicit $vgpr0, implicit killed $vgpr1, implicit killed $vgpr2, implicit killed $vgpr3, implicit-def $vgpr0
     ; CHECK: ADJCALLSTACKDOWN 0, 4, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr95
     ; CHECK: FLAT_STORE_DWORD undef renamable $vgpr0_vgpr1, killed renamable $vgpr0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
     ; CHECK: S_ENDPGM 0
     %0:vreg_128 = FLAT_LOAD_DWORDX4 undef %1:vreg_64, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16, addrspace 1)
     %2:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) @foo + 4, target-flags(amdgpu-rel32-hi) @foo + 4, implicit-def dead $scc
     ADJCALLSTACKUP 0, 0, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr95, implicit-def $scc

llvm/test/CodeGen/AMDGPU/wave32.ll

 ; Test save/restore of VGPR needed for SGPR spilling.
 
 ; GCN-LABEL: {{^}}callee_no_stack_with_call:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_waitcnt_vscnt
 
 ; GFX1064-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GFX1032-NEXT: s_or_saveexec_b32 [[COPY_EXEC0:s[0-9]]], -1{{$}}
-; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GCN-NEXT: v_nop
 ; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
 ; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC0]]
 
-; GCN-NEXT: v_writelane_b32 v32, s33, 2
+; GCN-NEXT: v_writelane_b32 v40, s33, 2
 ; GCN: s_mov_b32 s33, s32
 ; GFX1064: s_add_u32 s32, s32, 0x400
 ; GFX1032: s_add_u32 s32, s32, 0x200
 
 
-; GCN-DAG: v_writelane_b32 v32, s30, 0
-; GCN-DAG: v_writelane_b32 v32, s31, 1
+; GCN-DAG: v_writelane_b32 v40, s30, 0
+; GCN-DAG: v_writelane_b32 v40, s31, 1
 ; GCN: s_swappc_b64
-; GCN-DAG: v_readlane_b32 s4, v32, 0
-; GCN-DAG: v_readlane_b32 s5, v32, 1
+; GCN-DAG: v_readlane_b32 s4, v40, 0
+; GCN-DAG: v_readlane_b32 s5, v40, 1
 
 
 ; GFX1064: s_sub_u32 s32, s32, 0x400
 ; GFX1032: s_sub_u32 s32, s32, 0x200
-; GCN: v_readlane_b32 s33, v32, 2
+; GCN: v_readlane_b32 s33, v40, 2
 ; GFX1064: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; GFX1032: s_or_saveexec_b32 [[COPY_EXEC1:s[0-9]]], -1{{$}}
-; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GCN-NEXT: v_nop
 ; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
 ; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC1]]
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: s_setpc_b64
 define void @callee_no_stack_with_call() #1 {
   call void @external_void_func_void()
   ret void