[llvm-branch-commits] [llvm] [AMDGPU][GlobalISel] Fix immediate offset overflow when splitting s.buffer.load (PR #240932)

Krzysztof Drewniak via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Tue Nov 10 09:51:05 PST 2026


https://github.com/krzysz00 created https://github.com/llvm/llvm-project/pull/240932

When we're splitting a very large (>128 bit) s.buffer.load into
128-bit buffer loads, we add offsets to each split load. Currently,
when dealing with ptr.s.buffer.load, the code wouldn't subtract enough
from the immediate offset, so that when we added that +128 or +256,
we'd overflow the 12-bit immediate, causing silent overflow.

This commit fixes the issue by correctly calling the offset
splitter with a "alignment" that ensures that all those intermediate
loads don't overflow.

This matches the behavior in SDag and RegisterBankInfo.

Co-Authored-By: Claude Opus 5.5 <noreply at anthropic.com>

---

<sub>Stack created with <a href="https://github.com/github/gh-stack">GitHub Stacks CLI</a> • <a href="https://gh.io/stacks-feedback">Give Feedback 💬</a></sub>

>From 6b3194e79040da5d606dac66f9c56e1f8630c3a9 Mon Sep 17 00:00:00 2001
From: Krzysztof Drewniak <Krzysztof.Drewniak at amd.com>
Date: Mon, 9 Nov 2026 19:18:12 +0000
Subject: [PATCH] [AMDGPU][GlobalISel] Fix immediate offset overflow when
 splitting s.buffer.load

When we're splitting a very large (>128 bit) s.buffer.load into
128-bit buffer loads, we add offsets to each split load. Currently,
when dealing with ptr.s.buffer.load, the code wouldn't subtract enough
from the immediate offset, so that when we added that +128 or +256,
we'd overflow the 12-bit immediate, causing silent overflow.

This commit fixes the issue by correctly calling the offset
splitter with a "alignment" that ensures that all those intermediate
loads don't overflow.

This matches the behavior in SDag and RegisterBankInfo.

Co-Authored-By: Claude Opus 5.5 <noreply at anthropic.com>
---
 .../AMDGPU/AMDGPURegBankLegalizeHelper.cpp    |  5 +-
 .../llvm.amdgcn.ptr.s.buffer.load.ll          | 48 +++++++-------
 .../GlobalISel/llvm.amdgcn.s.buffer.load.ll   | 64 +++++++++----------
 .../regbankselect-amdgcn.ptr.s.buffer.load.ll | 36 +++++------
 .../regbankselect-amdgcn.s.buffer.load.ll     | 16 ++---
 5 files changed, 82 insertions(+), 87 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
index 07cce0d4e68e6f..1bacb10ba803bc 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -650,7 +650,10 @@ bool RegBankLegalizeHelper::lowerSBufToBuf(MachineInstr &MI,
   for (int I = 0; I < NumLoads; ++I)
     LoadParts.emplace_back(MRI.createVirtualRegister({VgprRB, Ty}));
   MachineMemOperand *OrigMMO = *MI.memoperands_begin();
-  const Align Alignment = OrigMMO->getAlign();
+  // Use the alignment to ensure that the required offsets will fit into the
+  // immediate offsets. This is unrelated to the alignment of the memory
+  // access.
+  const Align Alignment = NumLoads > 1 ? Align(16 * NumLoads) : Align(1);
   MachineFunction &MF = B.getMF();
   Register SOffset;
   Register VOffset;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ptr.s.buffer.load.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ptr.s.buffer.load.ll
index e4797843dd2288..00c272d8be599c 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ptr.s.buffer.load.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ptr.s.buffer.load.ll
@@ -3362,10 +3362,10 @@ define amdgpu_ps <8 x float> @ptr_s_buffer_load_v8f32_vgpr_offset_add_4068(ptr a
   ; GFX6-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
   ; GFX6-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX6-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4068
   ; GFX6-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 16, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
   ; GFX6-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7
   ; GFX6-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX6-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
@@ -3394,10 +3394,10 @@ define amdgpu_ps <8 x float> @ptr_s_buffer_load_v8f32_vgpr_offset_add_4068(ptr a
   ; GFX7-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
   ; GFX7-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4068
   ; GFX7-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 16, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
   ; GFX7-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX7-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
@@ -3427,9 +3427,9 @@ define amdgpu_ps <8 x float> @ptr_s_buffer_load_v8f32_vgpr_offset_add_4068(ptr a
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX8-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4064, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4080, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
   ; GFX8-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7
   ; GFX8-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX8-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
@@ -3784,12 +3784,12 @@ define amdgpu_ps <16 x float> @ptr_s_buffer_load_v16f32_vgpr_offset_add_4036(ptr
   ; GFX6-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
   ; GFX6-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX6-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4036
   ; GFX6-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4036, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4052, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 16, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 32, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 48, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
   ; GFX6-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_512 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7, [[BUFFER_LOAD_DWORDX4_OFFEN2]], %subreg.sub8_sub9_sub10_sub11, [[BUFFER_LOAD_DWORDX4_OFFEN3]], %subreg.sub12_sub13_sub14_sub15
   ; GFX6-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX6-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
@@ -3834,12 +3834,12 @@ define amdgpu_ps <16 x float> @ptr_s_buffer_load_v16f32_vgpr_offset_add_4036(ptr
   ; GFX7-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
   ; GFX7-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4036
   ; GFX7-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4036, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4052, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 16, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 32, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 48, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
   ; GFX7-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_512 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7, [[BUFFER_LOAD_DWORDX4_OFFEN2]], %subreg.sub8_sub9_sub10_sub11, [[BUFFER_LOAD_DWORDX4_OFFEN3]], %subreg.sub12_sub13_sub14_sub15
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX7-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
@@ -3885,11 +3885,11 @@ define amdgpu_ps <16 x float> @ptr_s_buffer_load_v16f32_vgpr_offset_add_4036(ptr
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX8-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4036, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4052, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4068, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4084, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4032, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN1:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4048, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN2:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4064, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN3:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4080, 0, 0, implicit $exec :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
   ; GFX8-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_512 = REG_SEQUENCE [[BUFFER_LOAD_DWORDX4_OFFEN]], %subreg.sub0_sub1_sub2_sub3, [[BUFFER_LOAD_DWORDX4_OFFEN1]], %subreg.sub4_sub5_sub6_sub7, [[BUFFER_LOAD_DWORDX4_OFFEN2]], %subreg.sub8_sub9_sub10_sub11, [[BUFFER_LOAD_DWORDX4_OFFEN3]], %subreg.sub12_sub13_sub14_sub15
   ; GFX8-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
   ; GFX8-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.buffer.load.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.buffer.load.ll
index eaa8fc972ddb31..6c5ccc831ee652 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.buffer.load.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.buffer.load.ll
@@ -2969,8 +2969,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_offset_add_4095(<4 x i32> inreg %
   ; GFX6-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX6-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
   ; GFX6-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
-  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
   ; GFX6-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
   ; GFX6-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
@@ -2984,8 +2984,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_offset_add_4095(<4 x i32> inreg %
   ; GFX7-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX7-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
   ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
-  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
   ; GFX7-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
   ; GFX7-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
@@ -2999,8 +2999,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_offset_add_4095(<4 x i32> inreg %
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
   ; GFX8-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 3
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4092, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
   ; GFX8-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
   ; GFX8-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
@@ -3144,8 +3144,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_offset_add_4096(<4 x i32> inreg %
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
   ; GFX8-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4092, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 1
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
   ; GFX8-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
   ; GFX8-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
@@ -4775,9 +4775,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX6-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
   ; GFX6-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
   ; GFX6-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
-  ; GFX6-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
-  ; GFX6-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; GFX6-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
   ; GFX6-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
   ; GFX6-NEXT: {{  $}}
   ; GFX6-NEXT: bb.2:
@@ -4788,19 +4786,19 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX6-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
   ; GFX6-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
   ; GFX6-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
-  ; GFX6-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
-  ; GFX6-NEXT:   [[COPY6:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
-  ; GFX6-NEXT:   [[COPY7:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
-  ; GFX6-NEXT:   [[COPY8:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
-  ; GFX6-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY7]], [[COPY5]], implicit $exec
-  ; GFX6-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY8]], [[COPY6]], implicit $exec
+  ; GFX6-NEXT:   [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; GFX6-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; GFX6-NEXT:   [[COPY6:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
+  ; GFX6-NEXT:   [[COPY7:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
+  ; GFX6-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY6]], [[COPY4]], implicit $exec
+  ; GFX6-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY7]], [[COPY5]], implicit $exec
   ; GFX6-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def dead $scc
   ; GFX6-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 [[S_AND_B64_]], implicit-def $exec, implicit-def dead $scc, implicit $exec
   ; GFX6-NEXT: {{  $}}
   ; GFX6-NEXT: bb.3:
   ; GFX6-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
   ; GFX6-NEXT: {{  $}}
-  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE1]], [[S_MOV_B32_1]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX6-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4095, align 1)
   ; GFX6-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def dead $scc
   ; GFX6-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
   ; GFX6-NEXT: {{  $}}
@@ -4810,7 +4808,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX6-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
   ; GFX6-NEXT: {{  $}}
   ; GFX6-NEXT: bb.5:
-  ; GFX6-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX6-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
   ; GFX6-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
   ; GFX7-LABEL: name: s_buffer_load_f32_vgpr_rsrc_offset_4095
@@ -4823,9 +4821,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
   ; GFX7-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
   ; GFX7-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
-  ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
-  ; GFX7-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; GFX7-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
   ; GFX7-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
   ; GFX7-NEXT: {{  $}}
   ; GFX7-NEXT: bb.2:
@@ -4836,19 +4832,19 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
   ; GFX7-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
   ; GFX7-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
-  ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
-  ; GFX7-NEXT:   [[COPY6:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
-  ; GFX7-NEXT:   [[COPY7:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
-  ; GFX7-NEXT:   [[COPY8:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
-  ; GFX7-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY7]], [[COPY5]], implicit $exec
-  ; GFX7-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY8]], [[COPY6]], implicit $exec
+  ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; GFX7-NEXT:   [[COPY6:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
+  ; GFX7-NEXT:   [[COPY7:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
+  ; GFX7-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY6]], [[COPY4]], implicit $exec
+  ; GFX7-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY7]], [[COPY5]], implicit $exec
   ; GFX7-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def dead $scc
   ; GFX7-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 [[S_AND_B64_]], implicit-def $exec, implicit-def dead $scc, implicit $exec
   ; GFX7-NEXT: {{  $}}
   ; GFX7-NEXT: bb.3:
   ; GFX7-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
   ; GFX7-NEXT: {{  $}}
-  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY4]], [[REG_SEQUENCE1]], [[S_MOV_B32_1]], 0, 0, 0, implicit $exec :: (dereferenceable invariant load (s32))
+  ; GFX7-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4095, align 1)
   ; GFX7-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def dead $scc
   ; GFX7-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
   ; GFX7-NEXT: {{  $}}
@@ -4858,7 +4854,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
   ; GFX7-NEXT: {{  $}}
   ; GFX7-NEXT: bb.5:
-  ; GFX7-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX7-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
   ; GFX7-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
   ; GFX8-LABEL: name: s_buffer_load_f32_vgpr_rsrc_offset_4095
@@ -4871,7 +4867,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX8-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 3
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
   ; GFX8-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
   ; GFX8-NEXT: {{  $}}
   ; GFX8-NEXT: bb.2:
@@ -4894,7 +4890,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX8-NEXT: bb.3:
   ; GFX8-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
   ; GFX8-NEXT: {{  $}}
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4092, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4095, align 1)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4095, align 1)
   ; GFX8-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def dead $scc
   ; GFX8-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
   ; GFX8-NEXT: {{  $}}
@@ -5108,7 +5104,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4096(<4 x i32> %rsrc)
   ; GFX8-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
   ; GFX8-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
   ; GFX8-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
-  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4
+  ; GFX8-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 1
   ; GFX8-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
   ; GFX8-NEXT: {{  $}}
   ; GFX8-NEXT: bb.2:
@@ -5131,7 +5127,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4096(<4 x i32> %rsrc)
   ; GFX8-NEXT: bb.3:
   ; GFX8-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
   ; GFX8-NEXT: {{  $}}
-  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4092, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4096)
+  ; GFX8-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET [[REG_SEQUENCE1]], [[S_MOV_B32_]], 4095, 0, 0, implicit $exec :: (dereferenceable invariant load (s32) from unknown-address + 4096)
   ; GFX8-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def dead $scc
   ; GFX8-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
   ; GFX8-NEXT: {{  $}}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.ptr.s.buffer.load.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.ptr.s.buffer.load.ll
index da5e6407add3cd..4479c111f68b08 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.ptr.s.buffer.load.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.ptr.s.buffer.load.ll
@@ -947,10 +947,9 @@ define amdgpu_ps <8 x float> @ptr_s_buffer_load_v8f32_vgpr_offset_add_4068(ptr a
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
   ; GFX7-NEXT:   [[ADD:%[0-9]+]]:vgpr(i32) = G_ADD [[COPY4]], [[COPY5]]
   ; GFX7-NEXT:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x i32>) = G_BUILD_VECTOR [[COPY]](i32), [[COPY1]](i32), [[COPY2]](i32), [[COPY3]](i32)
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4068, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4084, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 0, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 16, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
   ; GFX7-NEXT:   [[CONCAT_VECTORS:%[0-9]+]]:vgpr(<8 x f32>) = G_CONCAT_VECTORS [[AMDGPU_BUFFER_LOAD]](<4 x f32>), [[AMDGPU_BUFFER_LOAD1]](<4 x f32>)
   ; GFX7-NEXT:   [[UV:%[0-9]+]]:vgpr(f32), [[UV1:%[0-9]+]]:vgpr(f32), [[UV2:%[0-9]+]]:vgpr(f32), [[UV3:%[0-9]+]]:vgpr(f32), [[UV4:%[0-9]+]]:vgpr(f32), [[UV5:%[0-9]+]]:vgpr(f32), [[UV6:%[0-9]+]]:vgpr(f32), [[UV7:%[0-9]+]]:vgpr(f32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<8 x f32>)
   ; GFX7-NEXT:   $vgpr0 = COPY [[UV]](f32)
@@ -1161,12 +1160,11 @@ define amdgpu_ps <16 x float> @ptr_s_buffer_load_v16f32_vgpr_offset_add_4036(ptr
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
   ; GFX7-NEXT:   [[ADD:%[0-9]+]]:vgpr(i32) = G_ADD [[COPY4]], [[COPY5]]
   ; GFX7-NEXT:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x i32>) = G_BUILD_VECTOR [[COPY]](i32), [[COPY1]](i32), [[COPY2]](i32), [[COPY3]](i32)
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4036, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4052, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD2:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4068, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD3:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4084, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 0, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 16, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD2:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 32, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD3:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 48, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
   ; GFX7-NEXT:   [[CONCAT_VECTORS:%[0-9]+]]:vgpr(<16 x f32>) = G_CONCAT_VECTORS [[AMDGPU_BUFFER_LOAD]](<4 x f32>), [[AMDGPU_BUFFER_LOAD1]](<4 x f32>), [[AMDGPU_BUFFER_LOAD2]](<4 x f32>), [[AMDGPU_BUFFER_LOAD3]](<4 x f32>)
   ; GFX7-NEXT:   [[UV:%[0-9]+]]:vgpr(f32), [[UV1:%[0-9]+]]:vgpr(f32), [[UV2:%[0-9]+]]:vgpr(f32), [[UV3:%[0-9]+]]:vgpr(f32), [[UV4:%[0-9]+]]:vgpr(f32), [[UV5:%[0-9]+]]:vgpr(f32), [[UV6:%[0-9]+]]:vgpr(f32), [[UV7:%[0-9]+]]:vgpr(f32), [[UV8:%[0-9]+]]:vgpr(f32), [[UV9:%[0-9]+]]:vgpr(f32), [[UV10:%[0-9]+]]:vgpr(f32), [[UV11:%[0-9]+]]:vgpr(f32), [[UV12:%[0-9]+]]:vgpr(f32), [[UV13:%[0-9]+]]:vgpr(f32), [[UV14:%[0-9]+]]:vgpr(f32), [[UV15:%[0-9]+]]:vgpr(f32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<16 x f32>)
   ; GFX7-NEXT:   $vgpr0 = COPY [[UV]](f32)
@@ -1286,10 +1284,9 @@ define amdgpu_ps <8 x float> @ptr_s_buffer_load_v8f32_vgpr_offset_add_4092(ptr a
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
   ; GFX7-NEXT:   [[ADD:%[0-9]+]]:vgpr(i32) = G_ADD [[COPY4]], [[COPY5]]
   ; GFX7-NEXT:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x i32>) = G_BUILD_VECTOR [[COPY]](i32), [[COPY1]](i32), [[COPY2]](i32), [[COPY3]](i32)
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4092, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4108, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 0, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 16, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
   ; GFX7-NEXT:   [[CONCAT_VECTORS:%[0-9]+]]:vgpr(<8 x f32>) = G_CONCAT_VECTORS [[AMDGPU_BUFFER_LOAD]](<4 x f32>), [[AMDGPU_BUFFER_LOAD1]](<4 x f32>)
   ; GFX7-NEXT:   [[UV:%[0-9]+]]:vgpr(f32), [[UV1:%[0-9]+]]:vgpr(f32), [[UV2:%[0-9]+]]:vgpr(f32), [[UV3:%[0-9]+]]:vgpr(f32), [[UV4:%[0-9]+]]:vgpr(f32), [[UV5:%[0-9]+]]:vgpr(f32), [[UV6:%[0-9]+]]:vgpr(f32), [[UV7:%[0-9]+]]:vgpr(f32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<8 x f32>)
   ; GFX7-NEXT:   $vgpr0 = COPY [[UV]](f32)
@@ -1378,12 +1375,11 @@ define amdgpu_ps <16 x float> @ptr_s_buffer_load_v16f32_vgpr_offset_add_4092(ptr
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
   ; GFX7-NEXT:   [[ADD:%[0-9]+]]:vgpr(i32) = G_ADD [[COPY4]], [[COPY5]]
   ; GFX7-NEXT:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x i32>) = G_BUILD_VECTOR [[COPY]](i32), [[COPY1]](i32), [[COPY2]](i32), [[COPY3]](i32)
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4092, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4108, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD2:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4124, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD3:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4140, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 0, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD1:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 16, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 16, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD2:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 32, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 32, align 1, addrspace 8)
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD3:%[0-9]+]]:vgpr(<4 x f32>) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 48, 0, 0 :: (dereferenceable invariant load (s128) from %ir.rsrc + 48, align 1, addrspace 8)
   ; GFX7-NEXT:   [[CONCAT_VECTORS:%[0-9]+]]:vgpr(<16 x f32>) = G_CONCAT_VECTORS [[AMDGPU_BUFFER_LOAD]](<4 x f32>), [[AMDGPU_BUFFER_LOAD1]](<4 x f32>), [[AMDGPU_BUFFER_LOAD2]](<4 x f32>), [[AMDGPU_BUFFER_LOAD3]](<4 x f32>)
   ; GFX7-NEXT:   [[UV:%[0-9]+]]:vgpr(f32), [[UV1:%[0-9]+]]:vgpr(f32), [[UV2:%[0-9]+]]:vgpr(f32), [[UV3:%[0-9]+]]:vgpr(f32), [[UV4:%[0-9]+]]:vgpr(f32), [[UV5:%[0-9]+]]:vgpr(f32), [[UV6:%[0-9]+]]:vgpr(f32), [[UV7:%[0-9]+]]:vgpr(f32), [[UV8:%[0-9]+]]:vgpr(f32), [[UV9:%[0-9]+]]:vgpr(f32), [[UV10:%[0-9]+]]:vgpr(f32), [[UV11:%[0-9]+]]:vgpr(f32), [[UV12:%[0-9]+]]:vgpr(f32), [[UV13:%[0-9]+]]:vgpr(f32), [[UV14:%[0-9]+]]:vgpr(f32), [[UV15:%[0-9]+]]:vgpr(f32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<16 x f32>)
   ; GFX7-NEXT:   $vgpr0 = COPY [[UV]](f32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.s.buffer.load.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.s.buffer.load.ll
index 758f7d281f81ed..5a8141ea781bea 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.s.buffer.load.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.s.buffer.load.ll
@@ -1246,8 +1246,9 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_offset_add_4095(<4 x i32> inreg %
   ; GFX7-NEXT:   [[C:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 4095
   ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
   ; GFX7-NEXT:   [[ADD:%[0-9]+]]:vgpr(i32) = G_ADD [[COPY4]], [[COPY5]]
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(f32) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C1]](i32), [[COPY4]], [[C]], 0, 0, 0 :: (dereferenceable invariant load (s32))
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(f32) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 4095, 0, 0 :: (dereferenceable invariant load (s32))
   ; GFX7-NEXT:   [[COPY6:%[0-9]+]]:vgpr(f32) = COPY [[AMDGPU_BUFFER_LOAD]](f32)
   ; GFX7-NEXT:   $vgpr0 = COPY [[COPY6]](f32)
   ; GFX7-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
@@ -2191,9 +2192,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT:   [[COPY3:%[0-9]+]]:vgpr(i32) = COPY $vgpr3
   ; GFX7-NEXT:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x i32>) = G_BUILD_VECTOR [[COPY]](i32), [[COPY1]](i32), [[COPY2]](i32), [[COPY3]](i32)
   ; GFX7-NEXT:   [[C:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 4095
-  ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr(i32) = COPY [[C]](i32)
-  ; GFX7-NEXT:   [[C1:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
-  ; GFX7-NEXT:   [[C2:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[C1:%[0-9]+]]:vgpr(i32) = G_CONSTANT i32 0
+  ; GFX7-NEXT:   [[C2:%[0-9]+]]:sgpr(i32) = G_CONSTANT i32 0
   ; GFX7-NEXT:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
   ; GFX7-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
   ; GFX7-NEXT: {{  $}}
@@ -2215,7 +2215,7 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT: bb.3:
   ; GFX7-NEXT:   successors: %bb.4, %bb.2
   ; GFX7-NEXT: {{  $}}
-  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(f32) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR1]](<4 x i32>), [[C2]](i32), [[COPY4]], [[C1]], 0, 0, 0 :: (dereferenceable invariant load (s32))
+  ; GFX7-NEXT:   [[AMDGPU_BUFFER_LOAD:%[0-9]+]]:vgpr(f32) = G_AMDGPU_BUFFER_LOAD [[BUILD_VECTOR1]](<4 x i32>), [[C1]](i32), [[C1]], [[C2]], 4095, 0, 0 :: (dereferenceable invariant load (s32) from unknown-address + 4095, align 1)
   ; GFX7-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def dead $scc
   ; GFX7-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
   ; GFX7-NEXT: {{  $}}
@@ -2223,8 +2223,8 @@ define amdgpu_ps float @s_buffer_load_f32_vgpr_rsrc_offset_4095(<4 x i32> %rsrc)
   ; GFX7-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
   ; GFX7-NEXT: {{  $}}
   ; GFX7-NEXT: bb.5:
-  ; GFX7-NEXT:   [[COPY5:%[0-9]+]]:vgpr(f32) = COPY [[AMDGPU_BUFFER_LOAD]](f32)
-  ; GFX7-NEXT:   $vgpr0 = COPY [[COPY5]](f32)
+  ; GFX7-NEXT:   [[COPY4:%[0-9]+]]:vgpr(f32) = COPY [[AMDGPU_BUFFER_LOAD]](f32)
+  ; GFX7-NEXT:   $vgpr0 = COPY [[COPY4]](f32)
   ; GFX7-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ;
   ; GFX1200_1250-LABEL: name: s_buffer_load_f32_vgpr_rsrc_offset_4095



More information about the llvm-branch-commits mailing list