[llvm] 6aea630 - AMDGPU: Fix creating illegally typed readfirstlane in atomic optimizer (#128388)

via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 24 03:39:53 PST 2025


Author: Matt Arsenault
Date: 2025-02-24T18:39:49+07:00
New Revision: 6aea6308d125c15b05bf10766eb831f32140968a

URL: https://github.com/llvm/llvm-project/commit/6aea6308d125c15b05bf10766eb831f32140968a
DIFF: https://github.com/llvm/llvm-project/commit/6aea6308d125c15b05bf10766eb831f32140968a.diff

LOG: AMDGPU: Fix creating illegally typed readfirstlane in atomic optimizer (#128388)

We need to promote 8/16-bit cases to 32-bit. Unfortunately we are
missing demanded bits optimizations on readfirstlane, so we end up
emitting
an and instruction on the input. I'm also surprised this pass isn't
handling
half or bfloat yet.

Added: 
    llvm/test/CodeGen/AMDGPU/atomic-optimizer-promote-i8.ll

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
    llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
index 02f5ce2d18ff6..e46d0587e7943 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
@@ -898,8 +898,15 @@ void AMDGPUAtomicOptimizerImpl::optimizeAtomic(Instruction &I,
 
     // We need to broadcast the value who was the lowest active lane (the first
     // lane) to all other lanes in the wavefront.
-    Value *BroadcastI = nullptr;
-    BroadcastI = B.CreateIntrinsic(Ty, Intrinsic::amdgcn_readfirstlane, PHI);
+
+    Value *ReadlaneVal = PHI;
+    if (TyBitWidth < 32)
+      ReadlaneVal = B.CreateZExt(PHI, B.getInt32Ty());
+
+    Value *BroadcastI = B.CreateIntrinsic(
+        ReadlaneVal->getType(), Intrinsic::amdgcn_readfirstlane, ReadlaneVal);
+    if (TyBitWidth < 32)
+      BroadcastI = B.CreateTrunc(BroadcastI, Ty);
 
     // Now that we have the result of our single atomic operation, we need to
     // get our individual lane's slice into the result. We use the lane offset

diff  --git a/llvm/test/CodeGen/AMDGPU/atomic-optimizer-promote-i8.ll b/llvm/test/CodeGen/AMDGPU/atomic-optimizer-promote-i8.ll
new file mode 100644
index 0000000000000..d3e591634503f
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/atomic-optimizer-promote-i8.ll
@@ -0,0 +1,176 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -passes=amdgpu-atomic-optimizer %s | FileCheck %s
+
+define amdgpu_kernel void @uniform_or_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_or_i8(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i8 [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = lshr i64 [[TMP1]], 32
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; CHECK-NEXT:    [[TMP5:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP2]], i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP4]], i32 [[TMP5]])
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[TMP6]], 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[BB8:.*]], label %[[BB10:.*]]
+; CHECK:       [[BB8]]:
+; CHECK-NEXT:    [[TMP9:%.*]] = atomicrmw or ptr addrspace(1) [[UNIFORM_PTR]], i8 [[VAL]] monotonic, align 1
+; CHECK-NEXT:    br label %[[BB10]]
+; CHECK:       [[BB10]]:
+; CHECK-NEXT:    [[TMP11:%.*]] = phi i8 [ poison, [[TMP0:%.*]] ], [ [[TMP9]], %[[BB8]] ]
+; CHECK-NEXT:    [[TMP16:%.*]] = zext i8 [[TMP11]] to i32
+; CHECK-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TMP16]])
+; CHECK-NEXT:    [[TMP12:%.*]] = trunc i32 [[TMP17]] to i8
+; CHECK-NEXT:    [[TMP13:%.*]] = trunc i32 [[TMP6]] to i8
+; CHECK-NEXT:    [[TMP14:%.*]] = select i1 [[TMP7]], i8 0, i8 [[VAL]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or i8 [[TMP12]], [[TMP14]]
+; CHECK-NEXT:    store i8 [[TMP15]], ptr addrspace(1) [[RESULT]], align 1
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw or ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_add_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_add_i8(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i8 [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = lshr i64 [[TMP1]], 32
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; CHECK-NEXT:    [[TMP5:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP2]], i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP4]], i32 [[TMP5]])
+; CHECK-NEXT:    [[TMP7:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP1]])
+; CHECK-NEXT:    [[TMP8:%.*]] = trunc i64 [[TMP7]] to i8
+; CHECK-NEXT:    [[TMP9:%.*]] = mul i8 [[VAL]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i32 [[TMP6]], 0
+; CHECK-NEXT:    br i1 [[TMP10]], label %[[BB11:.*]], label %[[BB13:.*]]
+; CHECK:       [[BB11]]:
+; CHECK-NEXT:    [[TMP12:%.*]] = atomicrmw add ptr addrspace(1) [[UNIFORM_PTR]], i8 [[TMP9]] monotonic, align 1
+; CHECK-NEXT:    br label %[[BB13]]
+; CHECK:       [[BB13]]:
+; CHECK-NEXT:    [[TMP14:%.*]] = phi i8 [ poison, [[TMP0:%.*]] ], [ [[TMP12]], %[[BB11]] ]
+; CHECK-NEXT:    [[TMP19:%.*]] = zext i8 [[TMP14]] to i32
+; CHECK-NEXT:    [[TMP20:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TMP19]])
+; CHECK-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP20]] to i8
+; CHECK-NEXT:    [[TMP16:%.*]] = trunc i32 [[TMP6]] to i8
+; CHECK-NEXT:    [[TMP17:%.*]] = mul i8 [[VAL]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = add i8 [[TMP15]], [[TMP17]]
+; CHECK-NEXT:    store i8 [[TMP18]], ptr addrspace(1) [[RESULT]], align 1
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw add ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_xchg_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_xchg_i8(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i8 [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[RMW:%.*]] = atomicrmw xchg ptr addrspace(1) [[UNIFORM_PTR]], i8 [[VAL]] monotonic, align 1
+; CHECK-NEXT:    store i8 [[RMW]], ptr addrspace(1) [[RESULT]], align 1
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw xchg ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_or_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_or_i16(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i16 [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = lshr i64 [[TMP1]], 32
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; CHECK-NEXT:    [[TMP5:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP2]], i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP4]], i32 [[TMP5]])
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[TMP6]], 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[BB8:.*]], label %[[BB10:.*]]
+; CHECK:       [[BB8]]:
+; CHECK-NEXT:    [[TMP9:%.*]] = atomicrmw or ptr addrspace(1) [[UNIFORM_PTR]], i16 [[VAL]] monotonic, align 2
+; CHECK-NEXT:    br label %[[BB10]]
+; CHECK:       [[BB10]]:
+; CHECK-NEXT:    [[TMP11:%.*]] = phi i16 [ poison, [[TMP0:%.*]] ], [ [[TMP9]], %[[BB8]] ]
+; CHECK-NEXT:    [[TMP16:%.*]] = zext i16 [[TMP11]] to i32
+; CHECK-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TMP16]])
+; CHECK-NEXT:    [[TMP12:%.*]] = trunc i32 [[TMP17]] to i16
+; CHECK-NEXT:    [[TMP13:%.*]] = trunc i32 [[TMP6]] to i16
+; CHECK-NEXT:    [[TMP14:%.*]] = select i1 [[TMP7]], i16 0, i16 [[VAL]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or i16 [[TMP12]], [[TMP14]]
+; CHECK-NEXT:    store i16 [[TMP15]], ptr addrspace(1) [[RESULT]], align 2
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw or ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_add_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_add_i16(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i16 [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = lshr i64 [[TMP1]], 32
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; CHECK-NEXT:    [[TMP5:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP2]], i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP4]], i32 [[TMP5]])
+; CHECK-NEXT:    [[TMP7:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP1]])
+; CHECK-NEXT:    [[TMP8:%.*]] = trunc i64 [[TMP7]] to i16
+; CHECK-NEXT:    [[TMP9:%.*]] = mul i16 [[VAL]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i32 [[TMP6]], 0
+; CHECK-NEXT:    br i1 [[TMP10]], label %[[BB11:.*]], label %[[BB13:.*]]
+; CHECK:       [[BB11]]:
+; CHECK-NEXT:    [[TMP12:%.*]] = atomicrmw add ptr addrspace(1) [[UNIFORM_PTR]], i16 [[TMP9]] monotonic, align 2
+; CHECK-NEXT:    br label %[[BB13]]
+; CHECK:       [[BB13]]:
+; CHECK-NEXT:    [[TMP14:%.*]] = phi i16 [ poison, [[TMP0:%.*]] ], [ [[TMP12]], %[[BB11]] ]
+; CHECK-NEXT:    [[TMP19:%.*]] = zext i16 [[TMP14]] to i32
+; CHECK-NEXT:    [[TMP20:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TMP19]])
+; CHECK-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP20]] to i16
+; CHECK-NEXT:    [[TMP16:%.*]] = trunc i32 [[TMP6]] to i16
+; CHECK-NEXT:    [[TMP17:%.*]] = mul i16 [[VAL]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = add i16 [[TMP15]], [[TMP17]]
+; CHECK-NEXT:    store i16 [[TMP18]], ptr addrspace(1) [[RESULT]], align 2
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw add ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_xchg_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_xchg_i16(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], i16 [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[RMW:%.*]] = atomicrmw xchg ptr addrspace(1) [[UNIFORM_PTR]], i16 [[VAL]] monotonic, align 2
+; CHECK-NEXT:    store i16 [[RMW]], ptr addrspace(1) [[RESULT]], align 2
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw xchg ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_f16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, half %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_fadd_f16(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], half [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[RMW:%.*]] = atomicrmw fadd ptr addrspace(1) [[UNIFORM_PTR]], half [[VAL]] monotonic, align 2
+; CHECK-NEXT:    store half [[RMW]], ptr addrspace(1) [[RESULT]], align 2
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, half %val monotonic, align 2
+  store half %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_bf16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, bfloat %val) {
+; CHECK-LABEL: define amdgpu_kernel void @uniform_fadd_bf16(
+; CHECK-SAME: ptr addrspace(1) [[RESULT:%.*]], ptr addrspace(1) [[UNIFORM_PTR:%.*]], bfloat [[VAL:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[RMW:%.*]] = atomicrmw fadd ptr addrspace(1) [[UNIFORM_PTR]], bfloat [[VAL]] monotonic, align 2
+; CHECK-NEXT:    store bfloat [[RMW]], ptr addrspace(1) [[RESULT]], align 2
+; CHECK-NEXT:    ret void
+;
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, bfloat %val monotonic, align 2
+  store bfloat %rmw, ptr addrspace(1) %result
+  ret void
+}

diff  --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
index bc89a186db010..3737cc414c58f 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
@@ -7149,3 +7149,4244 @@ entry:
   store i64 %old, ptr addrspace(1) %out
   ret void
 }
+
+define amdgpu_kernel void @uniform_or_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; GFX7LESS-LABEL: uniform_or_i8:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT:    ; implicit-def: $vgpr0
+; GFX7LESS-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT:    s_cbranch_execz .LBB12_2
+; GFX7LESS-NEXT:  ; %bb.1:
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s8, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_and_b32 s7, s6, 0xff
+; GFX7LESS-NEXT:    s_lshl_b32 s7, s7, s2
+; GFX7LESS-NEXT:    s_mov_b32 s10, -1
+; GFX7LESS-NEXT:    s_mov_b32 s9, s3
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, s7
+; GFX7LESS-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s2, v0
+; GFX7LESS-NEXT:  .LBB12_2:
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7LESS-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX7LESS-NEXT:    v_cndmask_b32_e64 v0, v1, 0, vcc
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX7LESS-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_or_i8:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX8-NEXT:    ; implicit-def: $vgpr0
+; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX8-NEXT:    s_cbranch_execz .LBB12_2
+; GFX8-NEXT:  ; %bb.1:
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s8, s2, -4
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_mov_b32 s9, s3
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX8-NEXT:    s_and_b32 s3, s6, 0xff
+; GFX8-NEXT:    s_lshl_b32 s3, s3, s2
+; GFX8-NEXT:    s_mov_b32 s11, 0xf000
+; GFX8-NEXT:    s_mov_b32 s10, -1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s3
+; GFX8-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s2, v0
+; GFX8-NEXT:  .LBB12_2:
+; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX8-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s6
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX8-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_or_i8:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT:    ; implicit-def: $vgpr0
+; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT:    s_cbranch_execz .LBB12_2
+; GFX9-NEXT:  ; %bb.1:
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s8, s2, -4
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_mov_b32 s9, s3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    s_and_b32 s3, s6, 0xff
+; GFX9-NEXT:    s_lshl_b32 s3, s3, s2
+; GFX9-NEXT:    s_mov_b32 s11, 0xf000
+; GFX9-NEXT:    s_mov_b32 s10, -1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s2, v0
+; GFX9-NEXT:  .LBB12_2:
+; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s6
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_or_i8:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT:    ; implicit-def: $vgpr0
+; GFX1064-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1064-NEXT:  ; %bb.1:
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s7, s2, 3
+; GFX1064-NEXT:    s_and_b32 s8, s6, 0xff
+; GFX1064-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1064-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1064-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1064-NEXT:    s_and_b32 s8, s2, -4
+; GFX1064-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1064-NEXT:    s_mov_b32 s10, -1
+; GFX1064-NEXT:    s_mov_b32 s9, s3
+; GFX1064-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1064-NEXT:  .LBB12_2:
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1064-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1064-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1064-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_or_i8:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1032-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT:    ; implicit-def: $vgpr0
+; GFX1032-NEXT:    s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1032-NEXT:  ; %bb.1:
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s5, s2, 3
+; GFX1032-NEXT:    s_and_b32 s7, s6, 0xff
+; GFX1032-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX1032-NEXT:    s_and_b32 s8, s2, -4
+; GFX1032-NEXT:    s_lshl_b32 s7, s7, s5
+; GFX1032-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1032-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1032-NEXT:    s_mov_b32 s10, -1
+; GFX1032-NEXT:    s_mov_b32 s9, s3
+; GFX1032-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s5, v0
+; GFX1032-NEXT:  .LBB12_2:
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s4
+; GFX1032-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1032-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc_lo
+; GFX1032-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_or_i8:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1164-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1164-NEXT:    ; implicit-def: $vgpr0
+; GFX1164-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1164-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1164-NEXT:  ; %bb.1:
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s7, s2, 3
+; GFX1164-NEXT:    s_and_b32 s8, s6, 0xff
+; GFX1164-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1164-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1164-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1164-NEXT:    s_and_b32 s8, s2, -4
+; GFX1164-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1164-NEXT:    s_mov_b32 s10, -1
+; GFX1164-NEXT:    s_mov_b32 s9, s3
+; GFX1164-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1164-NEXT:  .LBB12_2:
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1164-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_or_i8:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s4, s[4:5], 0x34
+; GFX1132-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1132-NEXT:    ; implicit-def: $vgpr0
+; GFX1132-NEXT:    s_and_saveexec_b32 s5, vcc_lo
+; GFX1132-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1132-NEXT:  ; %bb.1:
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s6, s2, 3
+; GFX1132-NEXT:    s_and_b32 s7, s4, 0xff
+; GFX1132-NEXT:    s_lshl_b32 s6, s6, 3
+; GFX1132-NEXT:    s_and_b32 s8, s2, -4
+; GFX1132-NEXT:    s_lshl_b32 s7, s7, s6
+; GFX1132-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1132-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1132-NEXT:    s_mov_b32 s10, -1
+; GFX1132-NEXT:    s_mov_b32 s9, s3
+; GFX1132-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s6, v0
+; GFX1132-NEXT:  .LBB12_2:
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s5
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1132-NEXT:    v_cndmask_b32_e64 v0, s4, 0, vcc_lo
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_or_i8:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1264-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1264-NEXT:    ; implicit-def: $vgpr0
+; GFX1264-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1264-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1264-NEXT:  ; %bb.1:
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s7, s2, 3
+; GFX1264-NEXT:    s_and_b32 s8, s6, 0xff
+; GFX1264-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1264-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1264-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1264-NEXT:    s_and_b32 s8, s2, -4
+; GFX1264-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1264-NEXT:    s_mov_b32 s10, -1
+; GFX1264-NEXT:    s_mov_b32 s9, s3
+; GFX1264-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1264-NEXT:  .LBB12_2:
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1264-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1264-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_or_i8:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s4, s[4:5], 0x34
+; GFX1232-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1232-NEXT:    ; implicit-def: $vgpr0
+; GFX1232-NEXT:    s_and_saveexec_b32 s5, vcc_lo
+; GFX1232-NEXT:    s_cbranch_execz .LBB12_2
+; GFX1232-NEXT:  ; %bb.1:
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s6, s2, 3
+; GFX1232-NEXT:    s_and_b32 s7, s4, 0xff
+; GFX1232-NEXT:    s_lshl_b32 s6, s6, 3
+; GFX1232-NEXT:    s_and_b32 s8, s2, -4
+; GFX1232-NEXT:    s_lshl_b32 s7, s7, s6
+; GFX1232-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1232-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1232-NEXT:    s_mov_b32 s10, -1
+; GFX1232-NEXT:    s_mov_b32 s9, s3
+; GFX1232-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s6, v0
+; GFX1232-NEXT:  .LBB12_2:
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s5
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1232-NEXT:    v_cndmask_b32_e64 v0, s4, 0, vcc_lo
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw or ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_add_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; GFX7LESS-LABEL: uniform_add_i8:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_mov_b64 s[6:7], exec
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s10, s[4:5], 0xd
+; GFX7LESS-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s6, 0
+; GFX7LESS-NEXT:    v_mbcnt_hi_u32_b32_e32 v4, s7, v0
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX7LESS-NEXT:    ; implicit-def: $vgpr0
+; GFX7LESS-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX7LESS-NEXT:    s_cbranch_execz .LBB13_4
+; GFX7LESS-NEXT:  ; %bb.1:
+; GFX7LESS-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s11, s2, 3
+; GFX7LESS-NEXT:    s_mov_b64 s[2:3], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_mul_i32 s6, s10, s6
+; GFX7LESS-NEXT:    s_lshl_b32 s11, s11, 3
+; GFX7LESS-NEXT:    s_load_dword s15, s[4:5], 0x0
+; GFX7LESS-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX7LESS-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX7LESS-NEXT:    s_not_b32 s13, s12
+; GFX7LESS-NEXT:    s_lshl_b32 s14, s6, s11
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s15
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_add_i32_e32 v0, vcc, s14, v1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_and_b32_e32 v2, s13, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX7LESS-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX7LESS-NEXT:  .LBB13_4: ; %Flow
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX7LESS-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX7LESS-NEXT:    s_and_b32 s5, s10, 0xff
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, s4
+; GFX7LESS-NEXT:    v_mad_u32_u24 v0, s5, v4, v0
+; GFX7LESS-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_add_i8:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[6:7], exec
+; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX8-NEXT:    ; implicit-def: $vgpr0
+; GFX8-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX8-NEXT:    s_cbranch_execz .LBB13_4
+; GFX8-NEXT:  ; %bb.1:
+; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[6:7]
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_mul_i32 s6, s10, s4
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX8-NEXT:    s_and_b32 s2, s6, 0xff
+; GFX8-NEXT:    s_not_b32 s13, s12
+; GFX8-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX8-NEXT:    s_mov_b64 s[2:3], 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s14, v1
+; GFX8-NEXT:    v_and_b32_e32 v2, s13, v1
+; GFX8-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX8-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX8-NEXT:  .LBB13_4: ; %Flow
+; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX8-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_mad_u16 v0, s10, v4, v0
+; GFX8-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_add_i8:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[6:7], exec
+; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX9-NEXT:    ; implicit-def: $vgpr0
+; GFX9-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX9-NEXT:    s_cbranch_execz .LBB13_4
+; GFX9-NEXT:  ; %bb.1:
+; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[6:7]
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mul_i32 s6, s10, s4
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xff
+; GFX9-NEXT:    s_not_b32 s13, s12
+; GFX9-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX9-NEXT:    s_mov_b64 s[2:3], 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_add_u32_e32 v0, s14, v1
+; GFX9-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX9-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX9-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX9-NEXT:  .LBB13_4: ; %Flow
+; GFX9-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_mad_legacy_u16 v0, s10, v4, v0
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_add_i8:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1064-NEXT:    ; implicit-def: $vgpr0
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX1064-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX1064-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1064-NEXT:  ; %bb.1:
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX1064-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1064-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1064-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1064-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX1064-NEXT:    s_and_b32 s2, s2, 0xff
+; GFX1064-NEXT:    s_not_b32 s13, s12
+; GFX1064-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1064-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1064-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1064-NEXT:  .LBB13_4: ; %Flow
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1064-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1064-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_add_i8:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s8, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1032-NEXT:    s_mov_b32 s10, 0
+; GFX1032-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1032-NEXT:    ; implicit-def: $vgpr0
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v4
+; GFX1032-NEXT:    s_and_saveexec_b32 s9, vcc_lo
+; GFX1032-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1032-NEXT:  ; %bb.1:
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1032-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1032-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1032-NEXT:    s_not_b32 s11, s3
+; GFX1032-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1032-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s10
+; GFX1032-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1032-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:  .LBB13_4: ; %Flow
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1032-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_add_i8:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1164-NEXT:    s_mov_b64 s[8:9], exec
+; GFX1164-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1164-NEXT:    ; implicit-def: $vgpr0
+; GFX1164-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1164-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1164-NEXT:  ; %bb.1:
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1164-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1164-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1164-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1164-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX1164-NEXT:    s_and_b32 s2, s2, 0xff
+; GFX1164-NEXT:    s_not_b32 s13, s12
+; GFX1164-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1164-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1164-NEXT:    s_waitcnt_depctr 0xfff
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1164-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1164-NEXT:  .LBB13_4: ; %Flow
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_add_i8:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1132-NEXT:    s_mov_b32 s10, 0
+; GFX1132-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1132-NEXT:    s_mov_b32 s9, exec_lo
+; GFX1132-NEXT:    ; implicit-def: $vgpr0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1132-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1132-NEXT:  ; %bb.1:
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1132-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1132-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1132-NEXT:    s_not_b32 s11, s3
+; GFX1132-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1132-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s10
+; GFX1132-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1132-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:  .LBB13_4: ; %Flow
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_add_i8:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1264-NEXT:    s_mov_b64 s[8:9], exec
+; GFX1264-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1264-NEXT:    ; implicit-def: $vgpr0
+; GFX1264-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1264-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1264-NEXT:  ; %bb.1:
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1264-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1264-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1264-NEXT:    s_lshl_b32 s12, 0xff, s11
+; GFX1264-NEXT:    s_and_b32 s2, s2, 0xff
+; GFX1264-NEXT:    s_not_b32 s13, s12
+; GFX1264-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1264-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1264-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1264-NEXT:  .LBB13_4: ; %Flow
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1264-NEXT:    s_wait_alu 0xf1ff
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1264-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_add_i8:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1232-NEXT:    s_mov_b32 s10, 0
+; GFX1232-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1232-NEXT:    s_mov_b32 s9, exec_lo
+; GFX1232-NEXT:    ; implicit-def: $vgpr0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1232-NEXT:    s_cbranch_execz .LBB13_4
+; GFX1232-NEXT:  ; %bb.1:
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1232-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1232-NEXT:    s_not_b32 s11, s3
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:  .LBB13_2: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1232-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s10
+; GFX1232-NEXT:    s_cbranch_execnz .LBB13_2
+; GFX1232-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:  .LBB13_4: ; %Flow
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1232-NEXT:    s_wait_alu 0xf1ff
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw add ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_xchg_i8(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i8 %val) {
+; GFX7LESS-LABEL: uniform_xchg_i8:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_and_b32 s3, s6, 0xff
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0x0
+; GFX7LESS-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s11, 0xff, s10
+; GFX7LESS-NEXT:    s_lshl_b32 s2, s3, s10
+; GFX7LESS-NEXT:    s_not_b32 s3, s11
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, s3, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX7LESS-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_xchg_i8:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s2, 0xff, s8
+; GFX8-NEXT:    s_not_b32 s9, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, 0xff
+; GFX8-NEXT:    s_lshl_b32 s10, s2, s8
+; GFX8-NEXT:    s_mov_b64 s[2:3], 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_and_b32_e32 v0, s9, v1
+; GFX8-NEXT:    v_or_b32_e32 v0, s10, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX8-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_xchg_i8:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s2, 0xff, s8
+; GFX9-NEXT:    s_not_b32 s9, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xff
+; GFX9-NEXT:    s_lshl_b32 s10, s2, s8
+; GFX9-NEXT:    s_mov_b64 s[2:3], 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_and_b32_e32 v0, s9, v1
+; GFX9-NEXT:    v_or_b32_e32 v0, s10, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_xchg_i8:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX1064-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1064-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1064-NEXT:    s_lshl_b32 s2, 0xff, s8
+; GFX1064-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1064-NEXT:    s_not_b32 s9, s2
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_xchg_i8:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s9, 0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1032-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1032-NEXT:    s_not_b32 s3, s3
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_xchg_i8:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1164-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1164-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1164-NEXT:    s_lshl_b32 s2, 0xff, s8
+; GFX1164-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1164-NEXT:    s_not_b32 s9, s2
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_xchg_i8:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s9, 0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1132-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1132-NEXT:    s_not_b32 s3, s3
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_xchg_i8:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1264-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1264-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1264-NEXT:    s_lshl_b32 s2, 0xff, s8
+; GFX1264-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1264-NEXT:    s_not_b32 s9, s2
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_xchg_i8:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s9, 0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_and_b32 s6, s6, 0xff
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xff, s2
+; GFX1232-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1232-NEXT:    s_not_b32 s3, s3
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:  .LBB14_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_cbranch_execnz .LBB14_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b8 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw xchg ptr addrspace(1) %uniform.ptr, i8 %val monotonic, align 1
+  store i8 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_or_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; GFX7LESS-LABEL: uniform_or_i16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, exec_lo, 0
+; GFX7LESS-NEXT:    v_mbcnt_hi_u32_b32_e32 v1, exec_hi, v1
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GFX7LESS-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT:    s_cbranch_execz .LBB15_2
+; GFX7LESS-NEXT:  ; %bb.1:
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s8, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_and_b32 s7, s6, 0xffff
+; GFX7LESS-NEXT:    s_lshl_b32 s7, s7, s2
+; GFX7LESS-NEXT:    s_mov_b32 s10, -1
+; GFX7LESS-NEXT:    s_mov_b32 s9, s3
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, s7
+; GFX7LESS-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX7LESS-NEXT:    v_bfe_u32 v0, v0, s2, 16
+; GFX7LESS-NEXT:  .LBB15_2:
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7LESS-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX7LESS-NEXT:    v_cndmask_b32_e64 v0, v1, 0, vcc
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX7LESS-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_or_i16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX8-NEXT:    ; implicit-def: $vgpr0
+; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX8-NEXT:    s_cbranch_execz .LBB15_2
+; GFX8-NEXT:  ; %bb.1:
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s8, s2, -4
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_mov_b32 s9, s3
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX8-NEXT:    s_and_b32 s3, 0xffff, s6
+; GFX8-NEXT:    s_lshl_b32 s3, s3, s2
+; GFX8-NEXT:    s_mov_b32 s11, 0xf000
+; GFX8-NEXT:    s_mov_b32 s10, -1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s3
+; GFX8-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s2, v0
+; GFX8-NEXT:  .LBB15_2:
+; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX8-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s6
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX8-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_or_i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT:    ; implicit-def: $vgpr0
+; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT:    s_cbranch_execz .LBB15_2
+; GFX9-NEXT:  ; %bb.1:
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s8, s2, -4
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_mov_b32 s9, s3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    s_and_b32 s3, 0xffff, s6
+; GFX9-NEXT:    s_lshl_b32 s3, s3, s2
+; GFX9-NEXT:    s_mov_b32 s11, 0xf000
+; GFX9-NEXT:    s_mov_b32 s10, -1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s2, v0
+; GFX9-NEXT:  .LBB15_2:
+; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s6
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_or_b32_e32 v0, s4, v0
+; GFX9-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_or_i16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT:    ; implicit-def: $vgpr0
+; GFX1064-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1064-NEXT:  ; %bb.1:
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s7, s2, 3
+; GFX1064-NEXT:    s_and_b32 s8, 0xffff, s6
+; GFX1064-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1064-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1064-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1064-NEXT:    s_and_b32 s8, s2, -4
+; GFX1064-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1064-NEXT:    s_mov_b32 s10, -1
+; GFX1064-NEXT:    s_mov_b32 s9, s3
+; GFX1064-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1064-NEXT:  .LBB15_2:
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1064-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1064-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1064-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_or_i16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1032-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT:    ; implicit-def: $vgpr0
+; GFX1032-NEXT:    s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1032-NEXT:  ; %bb.1:
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s5, s2, 3
+; GFX1032-NEXT:    s_and_b32 s7, 0xffff, s6
+; GFX1032-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX1032-NEXT:    s_and_b32 s8, s2, -4
+; GFX1032-NEXT:    s_lshl_b32 s7, s7, s5
+; GFX1032-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1032-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1032-NEXT:    s_mov_b32 s10, -1
+; GFX1032-NEXT:    s_mov_b32 s9, s3
+; GFX1032-NEXT:    buffer_atomic_or v0, off, s[8:11], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s5, v0
+; GFX1032-NEXT:  .LBB15_2:
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s4
+; GFX1032-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1032-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc_lo
+; GFX1032-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_or_i16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1164-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1164-NEXT:    ; implicit-def: $vgpr0
+; GFX1164-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1164-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1164-NEXT:  ; %bb.1:
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s7, s2, 3
+; GFX1164-NEXT:    s_and_b32 s8, 0xffff, s6
+; GFX1164-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1164-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1164-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1164-NEXT:    s_and_b32 s8, s2, -4
+; GFX1164-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1164-NEXT:    s_mov_b32 s10, -1
+; GFX1164-NEXT:    s_mov_b32 s9, s3
+; GFX1164-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1164-NEXT:  .LBB15_2:
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1164-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_or_i16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s4, s[4:5], 0x34
+; GFX1132-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1132-NEXT:    ; implicit-def: $vgpr0
+; GFX1132-NEXT:    s_and_saveexec_b32 s5, vcc_lo
+; GFX1132-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1132-NEXT:  ; %bb.1:
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s6, s2, 3
+; GFX1132-NEXT:    s_and_b32 s7, 0xffff, s4
+; GFX1132-NEXT:    s_lshl_b32 s6, s6, 3
+; GFX1132-NEXT:    s_and_b32 s8, s2, -4
+; GFX1132-NEXT:    s_lshl_b32 s7, s7, s6
+; GFX1132-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1132-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1132-NEXT:    s_mov_b32 s10, -1
+; GFX1132-NEXT:    s_mov_b32 s9, s3
+; GFX1132-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s6, v0
+; GFX1132-NEXT:  .LBB15_2:
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s5
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1132-NEXT:    v_cndmask_b32_e64 v0, s4, 0, vcc_lo
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_or_i16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1264-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1264-NEXT:    ; implicit-def: $vgpr0
+; GFX1264-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1264-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1264-NEXT:  ; %bb.1:
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s7, s2, 3
+; GFX1264-NEXT:    s_and_b32 s8, 0xffff, s6
+; GFX1264-NEXT:    s_lshl_b32 s7, s7, 3
+; GFX1264-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1264-NEXT:    s_lshl_b32 s9, s8, s7
+; GFX1264-NEXT:    s_and_b32 s8, s2, -4
+; GFX1264-NEXT:    v_mov_b32_e32 v0, s9
+; GFX1264-NEXT:    s_mov_b32 s10, -1
+; GFX1264-NEXT:    s_mov_b32 s9, s3
+; GFX1264-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s7, v0
+; GFX1264-NEXT:  .LBB15_2:
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1264-NEXT:    v_cndmask_b32_e64 v0, s6, 0, vcc
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1264-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_or_i16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s4, s[4:5], 0x34
+; GFX1232-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1232-NEXT:    ; implicit-def: $vgpr0
+; GFX1232-NEXT:    s_and_saveexec_b32 s5, vcc_lo
+; GFX1232-NEXT:    s_cbranch_execz .LBB15_2
+; GFX1232-NEXT:  ; %bb.1:
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s6, s2, 3
+; GFX1232-NEXT:    s_and_b32 s7, 0xffff, s4
+; GFX1232-NEXT:    s_lshl_b32 s6, s6, 3
+; GFX1232-NEXT:    s_and_b32 s8, s2, -4
+; GFX1232-NEXT:    s_lshl_b32 s7, s7, s6
+; GFX1232-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1232-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1232-NEXT:    s_mov_b32 s10, -1
+; GFX1232-NEXT:    s_mov_b32 s9, s3
+; GFX1232-NEXT:    buffer_atomic_or_b32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s6, v0
+; GFX1232-NEXT:  .LBB15_2:
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s5
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1232-NEXT:    v_cndmask_b32_e64 v0, s4, 0, vcc_lo
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw or ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_add_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; GFX7LESS-LABEL: uniform_add_i16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_mov_b64 s[6:7], exec
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s10, s[4:5], 0xd
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, s6, 0
+; GFX7LESS-NEXT:    v_mbcnt_hi_u32_b32_e32 v4, s7, v1
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX7LESS-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX7LESS-NEXT:    s_cbranch_execz .LBB16_4
+; GFX7LESS-NEXT:  ; %bb.1:
+; GFX7LESS-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s11, s2, 3
+; GFX7LESS-NEXT:    s_mov_b64 s[2:3], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_mul_i32 s6, s10, s6
+; GFX7LESS-NEXT:    s_lshl_b32 s11, s11, 3
+; GFX7LESS-NEXT:    s_load_dword s15, s[4:5], 0x0
+; GFX7LESS-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX7LESS-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX7LESS-NEXT:    s_not_b32 s13, s12
+; GFX7LESS-NEXT:    s_lshl_b32 s14, s6, s11
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s15
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_add_i32_e32 v0, vcc, s14, v1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_and_b32_e32 v2, s13, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX7LESS-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX7LESS-NEXT:    v_bfe_u32 v0, v2, s11, 16
+; GFX7LESS-NEXT:  .LBB16_4: ; %Flow
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX7LESS-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX7LESS-NEXT:    s_and_b32 s5, s10, 0xffff
+; GFX7LESS-NEXT:    v_mov_b32_e32 v0, s4
+; GFX7LESS-NEXT:    v_mad_u32_u24 v0, s5, v4, v0
+; GFX7LESS-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_add_i16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[6:7], exec
+; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX8-NEXT:    ; implicit-def: $vgpr0
+; GFX8-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX8-NEXT:    s_cbranch_execz .LBB16_4
+; GFX8-NEXT:  ; %bb.1:
+; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[6:7]
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_mul_i32 s6, s10, s4
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX8-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX8-NEXT:    s_not_b32 s13, s12
+; GFX8-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX8-NEXT:    s_mov_b64 s[2:3], 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s14, v1
+; GFX8-NEXT:    v_and_b32_e32 v2, s13, v1
+; GFX8-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX8-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX8-NEXT:  .LBB16_4: ; %Flow
+; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX8-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_mad_u16 v0, s10, v4, v0
+; GFX8-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_add_i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[6:7], exec
+; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX9-NEXT:    ; implicit-def: $vgpr0
+; GFX9-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX9-NEXT:    s_cbranch_execz .LBB16_4
+; GFX9-NEXT:  ; %bb.1:
+; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[6:7]
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mul_i32 s6, s10, s4
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX9-NEXT:    s_not_b32 s13, s12
+; GFX9-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX9-NEXT:    s_mov_b64 s[2:3], 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_add_u32_e32 v0, s14, v1
+; GFX9-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX9-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX9-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX9-NEXT:  .LBB16_4: ; %Flow
+; GFX9-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_mad_legacy_u16 v0, s10, v4, v0
+; GFX9-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_add_i16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1064-NEXT:    ; implicit-def: $vgpr0
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX1064-NEXT:    s_and_saveexec_b64 s[8:9], vcc
+; GFX1064-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1064-NEXT:  ; %bb.1:
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX1064-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1064-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1064-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1064-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX1064-NEXT:    s_and_b32 s2, s2, 0xffff
+; GFX1064-NEXT:    s_not_b32 s13, s12
+; GFX1064-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1064-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1064-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1064-NEXT:  .LBB16_4: ; %Flow
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1064-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1064-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_add_i16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s8, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1032-NEXT:    s_mov_b32 s10, 0
+; GFX1032-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1032-NEXT:    ; implicit-def: $vgpr0
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v4
+; GFX1032-NEXT:    s_and_saveexec_b32 s9, vcc_lo
+; GFX1032-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1032-NEXT:  ; %bb.1:
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1032-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1032-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1032-NEXT:    s_not_b32 s11, s3
+; GFX1032-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1032-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s10
+; GFX1032-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1032-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:  .LBB16_4: ; %Flow
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1032-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_add_i16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1164-NEXT:    s_mov_b64 s[8:9], exec
+; GFX1164-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1164-NEXT:    ; implicit-def: $vgpr0
+; GFX1164-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1164-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1164-NEXT:  ; %bb.1:
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1164-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1164-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1164-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1164-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX1164-NEXT:    s_and_b32 s2, s2, 0xffff
+; GFX1164-NEXT:    s_not_b32 s13, s12
+; GFX1164-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1164-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1164-NEXT:    s_waitcnt_depctr 0xfff
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1164-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1164-NEXT:  .LBB16_4: ; %Flow
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_add_i16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1132-NEXT:    s_mov_b32 s10, 0
+; GFX1132-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1132-NEXT:    s_mov_b32 s9, exec_lo
+; GFX1132-NEXT:    ; implicit-def: $vgpr0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1132-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1132-NEXT:  ; %bb.1:
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1132-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1132-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1132-NEXT:    s_not_b32 s11, s3
+; GFX1132-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1132-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s10
+; GFX1132-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1132-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:  .LBB16_4: ; %Flow
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_add_i16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b64 s[6:7], exec
+; GFX1264-NEXT:    s_mov_b64 s[8:9], exec
+; GFX1264-NEXT:    v_mbcnt_lo_u32_b32 v0, s6, 0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_mbcnt_hi_u32_b32 v4, s7, v0
+; GFX1264-NEXT:    ; implicit-def: $vgpr0
+; GFX1264-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1264-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1264-NEXT:  ; %bb.1:
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1264-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
+; GFX1264-NEXT:    s_lshl_b32 s11, s2, 3
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_mul_i32 s2, s10, s6
+; GFX1264-NEXT:    s_lshl_b32 s12, 0xffff, s11
+; GFX1264-NEXT:    s_and_b32 s2, s2, 0xffff
+; GFX1264-NEXT:    s_not_b32 s13, s12
+; GFX1264-NEXT:    s_lshl_b32 s14, s2, s11
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_add_nc_u32_e32 v0, s14, v1
+; GFX1264-NEXT:    v_and_b32_e32 v0, s12, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s13, v0
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1264-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s11, v2
+; GFX1264-NEXT:  .LBB16_4: ; %Flow
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1264-NEXT:    s_wait_alu 0xf1ff
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1264-NEXT:    v_mad_u16 v0, s10, v4, s2
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_add_i16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s6, exec_lo
+; GFX1232-NEXT:    s_mov_b32 s10, 0
+; GFX1232-NEXT:    v_mbcnt_lo_u32_b32 v4, s6, 0
+; GFX1232-NEXT:    s_mov_b32 s9, exec_lo
+; GFX1232-NEXT:    ; implicit-def: $vgpr0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_cmpx_eq_u32_e32 0, v4
+; GFX1232-NEXT:    s_cbranch_execz .LBB16_4
+; GFX1232-NEXT:  ; %bb.1:
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1232-NEXT:    s_bcnt1_i32_b32 s6, s6
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_mul_i32 s6, s8, s6
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1232-NEXT:    s_not_b32 s11, s3
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_lshl_b32 s12, s6, s2
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:  .LBB16_2: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_add_nc_u32_e32 v0, s12, v1
+; GFX1232-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s11, v0
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s10, vcc_lo, s10
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s10
+; GFX1232-NEXT:    s_cbranch_execnz .LBB16_2
+; GFX1232-NEXT:  ; %bb.3: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s10
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:  .LBB16_4: ; %Flow
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1232-NEXT:    s_wait_alu 0xf1ff
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1232-NEXT:    v_mad_u16 v0, s8, v4, s2
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw add ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_xchg_i16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, i16 %val) {
+; GFX7LESS-LABEL: uniform_xchg_i16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_and_b32 s3, s6, 0xffff
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0x0
+; GFX7LESS-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s11, 0xffff, s10
+; GFX7LESS-NEXT:    s_lshl_b32 s2, s3, s10
+; GFX7LESS-NEXT:    s_not_b32 s3, s11
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, s3, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX7LESS-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_xchg_i16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX8-NEXT:    s_not_b32 s9, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s10, s2, s8
+; GFX8-NEXT:    s_mov_b64 s[2:3], 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_and_b32_e32 v0, s9, v1
+; GFX8-NEXT:    v_or_b32_e32 v0, s10, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX8-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_xchg_i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX9-NEXT:    s_not_b32 s9, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s10, s2, s8
+; GFX9-NEXT:    s_mov_b64 s[2:3], 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_and_b32_e32 v0, s9, v1
+; GFX9-NEXT:    v_or_b32_e32 v0, s10, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX9-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_xchg_i16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX1064-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1064-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1064-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1064-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1064-NEXT:    s_not_b32 s9, s2
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_xchg_i16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s9, 0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1032-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1032-NEXT:    s_not_b32 s3, s3
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_xchg_i16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1164-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1164-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1164-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1164-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1164-NEXT:    s_not_b32 s9, s2
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_xchg_i16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s9, 0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1132-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1132-NEXT:    s_not_b32 s3, s3
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_xchg_i16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1264-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1264-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1264-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1264-NEXT:    s_lshl_b32 s10, s6, s8
+; GFX1264-NEXT:    s_not_b32 s9, s2
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s9, s10
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_xchg_i16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s9, 0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_and_b32 s6, s6, 0xffff
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1232-NEXT:    s_lshl_b32 s8, s6, s2
+; GFX1232-NEXT:    s_not_b32 s3, s3
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:  .LBB17_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s3, s8
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_cbranch_execnz .LBB17_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw xchg ptr addrspace(1) %uniform.ptr, i16 %val monotonic, align 2
+  store i16 %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_f16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, half %val) {
+; GFX7LESS-LABEL: uniform_fadd_f16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v0, s6
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX7LESS-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; GFX7LESS-NEXT:    s_lshl_b32 s2, 0xffff, s10
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v4, v0
+; GFX7LESS-NEXT:    s_not_b32 s2, s2
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v1
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v0, v0
+; GFX7LESS-NEXT:    v_add_f32_e32 v0, v0, v4
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v0, s10, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_and_b32_e32 v2, s2, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX7LESS-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_fadd_f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s11, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[8:9], 0
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s2, 0xffff, s10
+; GFX8-NEXT:    s_not_b32 s2, s2
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s10, v1
+; GFX8-NEXT:    v_add_f16_e32 v0, s11, v0
+; GFX8-NEXT:    v_and_b32_e32 v2, s2, v1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, s10, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX8-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_fadd_f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s11, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[8:9], 0
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s2, 0xffff, s10
+; GFX9-NEXT:    s_not_b32 s2, s2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s10, v1
+; GFX9-NEXT:    v_add_f16_e32 v0, s11, v0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, s10, v0
+; GFX9-NEXT:    v_and_or_b32 v0, v1, s2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX9-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_fadd_f16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s8, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX1064-NEXT:    s_lshl_b32 s9, s2, 3
+; GFX1064-NEXT:    s_lshl_b32 s2, 0xffff, s9
+; GFX1064-NEXT:    s_not_b32 s10, s2
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s9, v1
+; GFX1064-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1064-NEXT:    v_lshlrev_b32_sdwa v0, s9, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s10, v0
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s9, v2
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_fadd_f16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s8, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s9, 0
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x0
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1032-NEXT:    s_not_b32 s3, s3
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s6
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v1
+; GFX1032-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1032-NEXT:    v_lshlrev_b32_sdwa v0, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s3, v0
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_fadd_f16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1164-NEXT:    s_lshl_b32 s9, s2, 3
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_lshl_b32 s2, 0xffff, s9
+; GFX1164-NEXT:    s_not_b32 s10, s2
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s9, v1
+; GFX1164-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1164-NEXT:    v_lshlrev_b32_e32 v0, s9, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s10, v0
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s9, v2
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_fadd_f16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s9, 0
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s6, s[4:5], 0x0
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1132-NEXT:    s_not_b32 s3, s3
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s6
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v1
+; GFX1132-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1132-NEXT:    v_lshlrev_b32_e32 v0, s2, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s3, v0
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_fadd_f16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s3, s[4:5], 0x0
+; GFX1264-NEXT:    s_lshl_b32 s9, s2, 3
+; GFX1264-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX1264-NEXT:    s_lshl_b32 s2, 0xffff, s9
+; GFX1264-NEXT:    s_not_b32 s10, s2
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s9, v1
+; GFX1264-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1264-NEXT:    v_lshlrev_b32_e32 v0, s9, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s10, v0
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_wait_alu 0xfffe
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s9, v2
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_fadd_f16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s9, 0
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s6, s[4:5], 0x0
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1232-NEXT:    s_not_b32 s3, s3
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s6
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:  .LBB18_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v1
+; GFX1232-NEXT:    v_add_f16_e32 v0, s8, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX1232-NEXT:    v_lshlrev_b32_e32 v0, s2, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s3, v0
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_cbranch_execnz .LBB18_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, half %val monotonic, align 2
+  store half %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_bf16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, bfloat %val) {
+; GFX7LESS-LABEL: uniform_fadd_bf16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7LESS-NEXT:    s_and_b32 s4, s2, -4
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:    s_and_b32 s2, s2, 3
+; GFX7LESS-NEXT:    s_lshl_b32 s10, s2, 3
+; GFX7LESS-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX7LESS-NEXT:    v_mul_f32_e64 v0, 1.0, s6
+; GFX7LESS-NEXT:    s_lshl_b32 s2, 0xffff, s10
+; GFX7LESS-NEXT:    v_and_b32_e32 v4, 0xffff0000, v0
+; GFX7LESS-NEXT:    s_not_b32 s2, s2
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v1
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX7LESS-NEXT:    v_add_f32_e32 v0, v0, v4
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v0, s10, v0
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_and_b32_e32 v2, s2, v1
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, v0
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_mov_b32_e32 v1, v2
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX7LESS-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_fadd_bf16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[8:9], 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s4, s2, -4
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX8-NEXT:    s_and_b32 s3, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s10, s3, 3
+; GFX8-NEXT:    s_lshl_b32 s3, 0xffff, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s6, 16
+; GFX8-NEXT:    s_not_b32 s3, s3
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s10
+; GFX8-NEXT:    v_lshrrev_b32_sdwa v3, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_add_f32_e32 v3, s2, v3
+; GFX8-NEXT:    v_bfe_u32 v4, v3, 16, 1
+; GFX8-NEXT:    v_add_u32_e32 v4, vcc, v4, v3
+; GFX8-NEXT:    v_add_u32_e32 v4, vcc, 0x7fff, v4
+; GFX8-NEXT:    v_or_b32_e32 v5, 0x400000, v3
+; GFX8-NEXT:    v_cmp_u_f32_e32 vcc, v3, v3
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX8-NEXT:    v_and_b32_e32 v2, s3, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX8-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_fadd_bf16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[8:9], 0
+; GFX9-NEXT:    s_movk_i32 s11, 0x7fff
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s4, s2, -4
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX9-NEXT:    s_and_b32 s3, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s10, s3, 3
+; GFX9-NEXT:    s_lshl_b32 s3, 0xffff, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s6, 16
+; GFX9-NEXT:    s_not_b32 s3, s3
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_lshrrev_b32_sdwa v0, s10, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_add_f32_e32 v0, s2, v0
+; GFX9-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX9-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX9-NEXT:    v_add3_u32 v2, v2, v0, s11
+; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s10, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_or_b32 v0, v1, s3, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, s10, v2
+; GFX9-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_fadd_bf16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_and_b32 s4, s2, -4
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_and_b32 s2, s2, 3
+; GFX1064-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1064-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1064-NEXT:    s_lshl_b32 s10, s6, 16
+; GFX1064-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_not_b32 s9, s2
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_lshrrev_b32_sdwa v0, s8, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX1064-NEXT:    v_add_f32_e32 v0, s10, v0
+; GFX1064-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1064-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1064-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
+; GFX1064-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1064-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX1064-NEXT:    v_lshlrev_b32_sdwa v0, s8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX1064-NEXT:    v_and_or_b32 v0, v1, s9, v0
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_fadd_bf16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s6, s[4:5], 0x34
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_and_b32 s4, s2, -4
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_and_b32 s2, s2, 3
+; GFX1032-NEXT:    s_load_dword s7, s[4:5], 0x0
+; GFX1032-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1032-NEXT:    s_lshl_b32 s9, s6, 16
+; GFX1032-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_not_b32 s8, s3
+; GFX1032-NEXT:    s_mov_b32 s3, 0
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_lshrrev_b32_sdwa v0, s2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX1032-NEXT:    v_add_f32_e32 v0, s9, v0
+; GFX1032-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1032-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1032-NEXT:    v_cmp_u_f32_e32 vcc_lo, v0, v0
+; GFX1032-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1032-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc_lo
+; GFX1032-NEXT:    v_lshlrev_b32_sdwa v0, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX1032-NEXT:    v_and_or_b32 v0, v1, s8, v0
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s3, vcc_lo, s3
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s3
+; GFX1032-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s3
+; GFX1032-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_fadd_bf16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_and_b32 s4, s2, -4
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_and_b32 s2, s2, 3
+; GFX1164-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1164-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1164-NEXT:    s_lshl_b32 s10, s6, 16
+; GFX1164-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_not_b32 s9, s2
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    .p2align 6
+; GFX1164-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s8, v1
+; GFX1164-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_add_f32_e32 v0, s10, v0
+; GFX1164-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1164-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1164-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1164-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX1164-NEXT:    v_lshlrev_b32_e32 v0, s8, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_and_or_b32 v0, v1, s9, v0
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_fadd_bf16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_and_b32 s4, s2, -4
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_and_b32 s2, s2, 3
+; GFX1132-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1132-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1132-NEXT:    s_lshl_b32 s9, s6, 16
+; GFX1132-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_not_b32 s8, s3
+; GFX1132-NEXT:    s_mov_b32 s3, 0
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:    .p2align 6
+; GFX1132-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v1
+; GFX1132-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_add_f32_e32 v0, s9, v0
+; GFX1132-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1132-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1132-NEXT:    v_cmp_u_f32_e32 vcc_lo, v0, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1132-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc_lo
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX1132-NEXT:    v_lshlrev_b32_e32 v0, s2, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_and_or_b32 v0, v1, s8, v0
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s3, vcc_lo, s3
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s3
+; GFX1132-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s3
+; GFX1132-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_fadd_bf16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_and_b32 s4, s2, -4
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_and_b32 s2, s2, 3
+; GFX1264-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1264-NEXT:    s_lshl_b32 s8, s2, 3
+; GFX1264-NEXT:    s_lshl_b32 s10, s6, 16
+; GFX1264-NEXT:    s_lshl_b32 s2, 0xffff, s8
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_not_b32 s9, s2
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s8, v1
+; GFX1264-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_add_f32_e32 v0, s10, v0
+; GFX1264-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1264-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1264-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1264-NEXT:    s_wait_alu 0xfffd
+; GFX1264-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX1264-NEXT:    v_lshlrev_b32_e32 v0, s8, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_and_or_b32 v0, v1, s9, v0
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    v_lshrrev_b32_e32 v0, s8, v2
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_fadd_bf16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s6, s[4:5], 0x34
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_and_b32 s4, s2, -4
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_and_b32 s2, s2, 3
+; GFX1232-NEXT:    s_load_b32 s7, s[4:5], 0x0
+; GFX1232-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX1232-NEXT:    s_lshl_b32 s9, s6, 16
+; GFX1232-NEXT:    s_lshl_b32 s3, 0xffff, s2
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_not_b32 s8, s3
+; GFX1232-NEXT:    s_mov_b32 s3, 0
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s7
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:  .LBB19_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v1
+; GFX1232-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_add_f32_e32 v0, s9, v0
+; GFX1232-NEXT:    v_bfe_u32 v2, v0, 16, 1
+; GFX1232-NEXT:    v_or_b32_e32 v3, 0x400000, v0
+; GFX1232-NEXT:    v_cmp_u_f32_e32 vcc_lo, v0, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_add3_u32 v2, v2, v0, 0x7fff
+; GFX1232-NEXT:    s_wait_alu 0xfffd
+; GFX1232-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc_lo
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX1232-NEXT:    v_lshlrev_b32_e32 v0, s2, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_and_or_b32 v0, v1, s8, v0
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s3, vcc_lo, s3
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s3
+; GFX1232-NEXT:    s_cbranch_execnz .LBB19_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s3
+; GFX1232-NEXT:    v_lshrrev_b32_e32 v0, s2, v2
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b16 v0, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, bfloat %val monotonic, align 2
+  store bfloat %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_v2f16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, <2 x half> %val) {
+; GFX7LESS-LABEL: uniform_fadd_v2f16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_lshr_b32 s4, s6, 16
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v0, s6
+; GFX7LESS-NEXT:    s_load_dword s5, s[2:3], 0x0
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v1, s4
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v1, v1
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v0, v0
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v1, v1
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v2, s5
+; GFX7LESS-NEXT:    s_lshr_b32 s4, s5, 16
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v3, s4
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:    s_mov_b32 s4, s2
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v4, v3
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v5, v2
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7LESS-NEXT:    v_add_f32_e32 v4, v4, v1
+; GFX7LESS-NEXT:    v_add_f32_e32 v5, v5, v0
+; GFX7LESS-NEXT:    v_or_b32_e32 v3, v2, v3
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v2, v4
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v4, v5
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7LESS-NEXT:    v_or_b32_e32 v2, v4, v2
+; GFX7LESS-NEXT:    v_mov_b32_e32 v5, v3
+; GFX7LESS-NEXT:    v_mov_b32_e32 v4, v2
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[4:5], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v3, 16, v4
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v2, v4
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_cvt_f32_f16_e32 v3, v3
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v0, v3
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX7LESS-NEXT:    v_cvt_f16_f32_e32 v1, v2
+; GFX7LESS-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7LESS-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_fadd_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[8:9], 0
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s5, s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s11, s10, 16
+; GFX8-NEXT:    s_mov_b32 s4, s2
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    s_mov_b32 s5, s3
+; GFX8-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s11
+; GFX8-NEXT:    v_add_f16_e32 v2, s10, v1
+; GFX8-NEXT:    v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX8-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8-NEXT:    s_mov_b32 s2, -1
+; GFX8-NEXT:    buffer_store_dword v2, off, s[0:3], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_fadd_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[8:9], 0
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s5, s[2:3], 0x0
+; GFX9-NEXT:    s_mov_b32 s4, s2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    s_mov_b32 s5, s3
+; GFX9-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_pk_add_f16 v0, v1, s10
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    buffer_store_dword v2, off, s[0:3], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_fadd_v2f16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s10, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b64 s[8:9], 0
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_load_dword s4, s[2:3], 0x0
+; GFX1064-NEXT:    s_mov_b32 s5, s3
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1064-NEXT:    s_mov_b32 s4, s2
+; GFX1064-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_pk_add_f16 v0, v1, s10
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s2, -1
+; GFX1064-NEXT:    buffer_store_dword v2, off, s[0:3], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_fadd_v2f16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s8, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s9, 0
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_load_dword s4, s[2:3], 0x0
+; GFX1032-NEXT:    s_mov_b32 s5, s3
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1032-NEXT:    s_mov_b32 s4, s2
+; GFX1032-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_pk_add_f16 v0, v1, s8
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1032-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s2, -1
+; GFX1032-NEXT:    buffer_store_dword v2, off, s[0:3], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_fadd_v2f16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b64 s[8:9], 0
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_load_b32 s4, s[2:3], 0x0
+; GFX1164-NEXT:    s_mov_b32 s5, s3
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1164-NEXT:    s_mov_b32 s4, s2
+; GFX1164-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_pk_add_f16 v0, v1, s10
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[8:9]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1164-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s2, -1
+; GFX1164-NEXT:    buffer_store_b32 v2, off, s[0:3], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_fadd_v2f16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s9, 0
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_load_b32 s4, s[2:3], 0x0
+; GFX1132-NEXT:    s_mov_b32 s5, s3
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1132-NEXT:    s_mov_b32 s4, s2
+; GFX1132-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_pk_add_f16 v0, v1, s8
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1132-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s2, -1
+; GFX1132-NEXT:    buffer_store_b32 v2, off, s[0:3], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_fadd_v2f16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s10, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b64 s[8:9], 0
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_load_b32 s4, s[2:3], 0x0
+; GFX1264-NEXT:    s_mov_b32 s5, s3
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1264-NEXT:    s_mov_b32 s4, s2
+; GFX1264-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_pk_add_f16 v0, v1, s10
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX1264-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[8:9]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX1264-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s2, -1
+; GFX1264-NEXT:    buffer_store_b32 v2, off, s[0:3], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_fadd_v2f16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s8, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s9, 0
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_load_b32 s4, s[2:3], 0x0
+; GFX1232-NEXT:    s_mov_b32 s5, s3
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1232-NEXT:    s_mov_b32 s4, s2
+; GFX1232-NEXT:  .LBB20_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_pk_add_f16 v0, v1, s8
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s9, vcc_lo, s9
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_cbranch_execnz .LBB20_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s9
+; GFX1232-NEXT:    s_mov_b32 s3, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s2, -1
+; GFX1232-NEXT:    buffer_store_b32 v2, off, s[0:3], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, <2 x half> %val monotonic, align 4
+  store <2 x half> %rmw, ptr addrspace(1) %result
+  ret void
+}
+
+define amdgpu_kernel void @uniform_fadd_v2bf16(ptr addrspace(1) %result, ptr addrspace(1) %uniform.ptr, <2 x bfloat> %val) {
+; GFX7LESS-LABEL: uniform_fadd_v2bf16:
+; GFX7LESS:       ; %bb.0:
+; GFX7LESS-NEXT:    s_load_dword s6, s[4:5], 0xd
+; GFX7LESS-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
+; GFX7LESS-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7LESS-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s6, 0xffff0000
+; GFX7LESS-NEXT:    s_lshl_b32 s5, s6, 16
+; GFX7LESS-NEXT:    s_load_dword s6, s[2:3], 0x0
+; GFX7LESS-NEXT:    v_mul_f32_e64 v0, 1.0, s5
+; GFX7LESS-NEXT:    v_mul_f32_e64 v1, 1.0, s4
+; GFX7LESS-NEXT:    v_and_b32_e32 v0, 0xffff0000, v0
+; GFX7LESS-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT:    s_and_b32 s4, s6, 0xffff0000
+; GFX7LESS-NEXT:    s_lshl_b32 s5, s6, 16
+; GFX7LESS-NEXT:    v_and_b32_e32 v1, 0xffff0000, v1
+; GFX7LESS-NEXT:    v_mov_b32_e32 v3, s5
+; GFX7LESS-NEXT:    v_mov_b32_e32 v2, s4
+; GFX7LESS-NEXT:    s_mov_b32 s6, -1
+; GFX7LESS-NEXT:    s_mov_b32 s4, s2
+; GFX7LESS-NEXT:    s_mov_b32 s5, s3
+; GFX7LESS-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX7LESS-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT:    v_mul_f32_e32 v3, 1.0, v3
+; GFX7LESS-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX7LESS-NEXT:    s_waitcnt expcnt(0)
+; GFX7LESS-NEXT:    v_and_b32_e32 v4, 0xffff0000, v3
+; GFX7LESS-NEXT:    v_and_b32_e32 v5, 0xffff0000, v2
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; GFX7LESS-NEXT:    v_add_f32_e32 v4, v4, v0
+; GFX7LESS-NEXT:    v_add_f32_e32 v5, v5, v1
+; GFX7LESS-NEXT:    v_alignbit_b32 v3, v2, v3, 16
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v2, 16, v5
+; GFX7LESS-NEXT:    v_alignbit_b32 v2, v2, v4, 16
+; GFX7LESS-NEXT:    v_mov_b32_e32 v5, v3
+; GFX7LESS-NEXT:    v_mov_b32_e32 v4, v2
+; GFX7LESS-NEXT:    buffer_atomic_cmpswap v[4:5], off, s[4:7], 0 glc
+; GFX7LESS-NEXT:    s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
+; GFX7LESS-NEXT:    v_and_b32_e32 v2, 0xffff0000, v4
+; GFX7LESS-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
+; GFX7LESS-NEXT:    v_lshlrev_b32_e32 v3, 16, v4
+; GFX7LESS-NEXT:    s_andn2_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX7LESS-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT:    s_or_b64 exec, exec, s[8:9]
+; GFX7LESS-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT:    s_mov_b32 s2, -1
+; GFX7LESS-NEXT:    v_mul_f32_e32 v0, 1.0, v2
+; GFX7LESS-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX7LESS-NEXT:    v_mul_f32_e32 v1, 1.0, v3
+; GFX7LESS-NEXT:    v_alignbit_b32 v0, v0, v1, 16
+; GFX7LESS-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7LESS-NEXT:    s_endpgm
+;
+; GFX8-LABEL: uniform_fadd_v2bf16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x24
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x34
+; GFX8-NEXT:    s_mov_b64 s[2:3], 0
+; GFX8-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8-NEXT:    s_mov_b32 s6, -1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s1, s[10:11], 0x0
+; GFX8-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX8-NEXT:    s_and_b32 s13, s0, 0xffff0000
+; GFX8-NEXT:    s_mov_b32 s4, s10
+; GFX8-NEXT:    s_mov_b32 s5, s11
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX8-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX8-NEXT:    v_add_f32_e32 v0, s12, v0
+; GFX8-NEXT:    v_add_f32_e32 v2, s13, v2
+; GFX8-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX8-NEXT:    v_bfe_u32 v5, v2, 16, 1
+; GFX8-NEXT:    v_add_u32_e32 v3, vcc, v3, v0
+; GFX8-NEXT:    v_add_u32_e32 v5, vcc, v5, v2
+; GFX8-NEXT:    v_add_u32_e32 v3, vcc, 0x7fff, v3
+; GFX8-NEXT:    v_add_u32_e32 v5, vcc, 0x7fff, v5
+; GFX8-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX8-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; GFX8-NEXT:    v_or_b32_e32 v4, 0x400000, v0
+; GFX8-NEXT:    v_cmp_u_f32_e64 s[0:1], v0, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v5, v6, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v3, v4, s[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; GFX8-NEXT:    v_alignbit_b32 v0, v2, v0, 16
+; GFX8-NEXT:    v_mov_b32_e32 v3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, v0
+; GFX8-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v1, v2
+; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX8-NEXT:    s_mov_b32 s11, 0xf000
+; GFX8-NEXT:    s_mov_b32 s10, -1
+; GFX8-NEXT:    buffer_store_dword v2, off, s[8:11], 0
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: uniform_fadd_v2bf16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x24
+; GFX9-NEXT:    s_load_dword s0, s[4:5], 0x34
+; GFX9-NEXT:    s_mov_b64 s[2:3], 0
+; GFX9-NEXT:    s_movk_i32 s12, 0x7fff
+; GFX9-NEXT:    s_mov_b32 s13, 0x7060302
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s1, s[10:11], 0x0
+; GFX9-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9-NEXT:    s_mov_b32 s6, -1
+; GFX9-NEXT:    s_lshl_b32 s14, s0, 16
+; GFX9-NEXT:    s_and_b32 s15, s0, 0xffff0000
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    s_mov_b32 s4, s10
+; GFX9-NEXT:    s_mov_b32 s5, s11
+; GFX9-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX9-NEXT:    v_add_f32_e32 v0, s14, v0
+; GFX9-NEXT:    v_add_f32_e32 v2, s15, v2
+; GFX9-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX9-NEXT:    v_bfe_u32 v5, v2, 16, 1
+; GFX9-NEXT:    v_or_b32_e32 v4, 0x400000, v0
+; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX9-NEXT:    v_add3_u32 v3, v3, v0, s12
+; GFX9-NEXT:    v_add3_u32 v5, v5, v2, s12
+; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; GFX9-NEXT:    v_cmp_u_f32_e64 s[0:1], v0, v0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v3, v4, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v5, v6, vcc
+; GFX9-NEXT:    v_perm_b32 v0, v2, v0, s13
+; GFX9-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, v0
+; GFX9-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX9-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
+; GFX9-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX9-NEXT:    s_mov_b32 s11, 0xf000
+; GFX9-NEXT:    s_mov_b32 s10, -1
+; GFX9-NEXT:    buffer_store_dword v2, off, s[8:11], 0
+; GFX9-NEXT:    s_endpgm
+;
+; GFX1064-LABEL: uniform_fadd_v2bf16:
+; GFX1064:       ; %bb.0:
+; GFX1064-NEXT:    s_clause 0x1
+; GFX1064-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x24
+; GFX1064-NEXT:    s_load_dword s0, s[4:5], 0x34
+; GFX1064-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1064-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s6, -1
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    s_load_dword s1, s[10:11], 0x0
+; GFX1064-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX1064-NEXT:    s_and_b32 s13, s0, 0xffff0000
+; GFX1064-NEXT:    s_mov_b32 s4, s10
+; GFX1064-NEXT:    s_mov_b32 s5, s11
+; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT:    v_mov_b32_e32 v1, s1
+; GFX1064-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1064-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1064-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1064-NEXT:    v_add_f32_e32 v0, s12, v0
+; GFX1064-NEXT:    v_add_f32_e32 v2, s13, v2
+; GFX1064-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1064-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1064-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1064-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1064-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; GFX1064-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1064-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1064-NEXT:    v_cmp_u_f32_e64 s[0:1], v0, v0
+; GFX1064-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc
+; GFX1064-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s[0:1]
+; GFX1064-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1064-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1064-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1064-NEXT:    s_waitcnt vmcnt(0)
+; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1064-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1064-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1064-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1064-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1064-NEXT:    s_mov_b32 s10, -1
+; GFX1064-NEXT:    buffer_store_dword v2, off, s[8:11], 0
+; GFX1064-NEXT:    s_endpgm
+;
+; GFX1032-LABEL: uniform_fadd_v2bf16:
+; GFX1032:       ; %bb.0:
+; GFX1032-NEXT:    s_clause 0x1
+; GFX1032-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x24
+; GFX1032-NEXT:    s_load_dword s0, s[4:5], 0x34
+; GFX1032-NEXT:    s_mov_b32 s1, 0
+; GFX1032-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s6, -1
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    s_load_dword s4, s[10:11], 0x0
+; GFX1032-NEXT:    s_lshl_b32 s2, s0, 16
+; GFX1032-NEXT:    s_and_b32 s3, s0, 0xffff0000
+; GFX1032-NEXT:    s_mov_b32 s5, s11
+; GFX1032-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1032-NEXT:    s_mov_b32 s4, s10
+; GFX1032-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1032-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1032-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1032-NEXT:    v_add_f32_e32 v0, s2, v0
+; GFX1032-NEXT:    v_add_f32_e32 v2, s3, v2
+; GFX1032-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1032-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1032-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1032-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1032-NEXT:    v_cmp_u_f32_e32 vcc_lo, v2, v2
+; GFX1032-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1032-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1032-NEXT:    v_cmp_u_f32_e64 s0, v0, v0
+; GFX1032-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc_lo
+; GFX1032-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s0
+; GFX1032-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1032-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1032-NEXT:    buffer_atomic_cmpswap v[2:3], off, s[4:7], 0 glc
+; GFX1032-NEXT:    s_waitcnt vmcnt(0)
+; GFX1032-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1032-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1032-NEXT:    s_or_b32 s1, vcc_lo, s1
+; GFX1032-NEXT:    s_andn2_b32 exec_lo, exec_lo, s1
+; GFX1032-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1032-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT:    s_or_b32 exec_lo, exec_lo, s1
+; GFX1032-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1032-NEXT:    s_mov_b32 s10, -1
+; GFX1032-NEXT:    buffer_store_dword v2, off, s[8:11], 0
+; GFX1032-NEXT:    s_endpgm
+;
+; GFX1164-LABEL: uniform_fadd_v2bf16:
+; GFX1164:       ; %bb.0:
+; GFX1164-NEXT:    s_clause 0x1
+; GFX1164-NEXT:    s_load_b128 s[8:11], s[4:5], 0x24
+; GFX1164-NEXT:    s_load_b32 s0, s[4:5], 0x34
+; GFX1164-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1164-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s6, -1
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    s_load_b32 s1, s[10:11], 0x0
+; GFX1164-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX1164-NEXT:    s_and_b32 s13, s0, 0xffff0000
+; GFX1164-NEXT:    s_mov_b32 s4, s10
+; GFX1164-NEXT:    s_mov_b32 s5, s11
+; GFX1164-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT:    v_mov_b32_e32 v1, s1
+; GFX1164-NEXT:    s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT:    .p2align 6
+; GFX1164-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1164-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1164-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1164-NEXT:    v_add_f32_e32 v0, s12, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_add_f32_e32 v2, s13, v2
+; GFX1164-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1164-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1164-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1164-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; GFX1164-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1164-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1164-NEXT:    v_cmp_u_f32_e64 s[0:1], v0, v0
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc
+; GFX1164-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s[0:1]
+; GFX1164-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1164-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1164-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1164-NEXT:    s_waitcnt vmcnt(0)
+; GFX1164-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1164-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1164-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1164-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT:    s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1164-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1164-NEXT:    s_mov_b32 s10, -1
+; GFX1164-NEXT:    buffer_store_b32 v2, off, s[8:11], 0
+; GFX1164-NEXT:    s_endpgm
+;
+; GFX1132-LABEL: uniform_fadd_v2bf16:
+; GFX1132:       ; %bb.0:
+; GFX1132-NEXT:    s_clause 0x1
+; GFX1132-NEXT:    s_load_b128 s[8:11], s[4:5], 0x24
+; GFX1132-NEXT:    s_load_b32 s0, s[4:5], 0x34
+; GFX1132-NEXT:    s_mov_b32 s1, 0
+; GFX1132-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s6, -1
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    s_load_b32 s4, s[10:11], 0x0
+; GFX1132-NEXT:    s_lshl_b32 s2, s0, 16
+; GFX1132-NEXT:    s_and_b32 s3, s0, 0xffff0000
+; GFX1132-NEXT:    s_mov_b32 s5, s11
+; GFX1132-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1132-NEXT:    s_mov_b32 s4, s10
+; GFX1132-NEXT:    s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT:    .p2align 6
+; GFX1132-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1132-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1132-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1132-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1132-NEXT:    v_add_f32_e32 v0, s2, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-NEXT:    v_add_f32_e32 v2, s3, v2
+; GFX1132-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1132-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1132-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1132-NEXT:    v_cmp_u_f32_e32 vcc_lo, v2, v2
+; GFX1132-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1132-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1132-NEXT:    v_cmp_u_f32_e64 s0, v0, v0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc_lo
+; GFX1132-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s0
+; GFX1132-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1132-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], 0 glc
+; GFX1132-NEXT:    s_waitcnt vmcnt(0)
+; GFX1132-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1132-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1132-NEXT:    s_or_b32 s1, vcc_lo, s1
+; GFX1132-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
+; GFX1132-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1132-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT:    s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s1
+; GFX1132-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1132-NEXT:    s_mov_b32 s10, -1
+; GFX1132-NEXT:    buffer_store_b32 v2, off, s[8:11], 0
+; GFX1132-NEXT:    s_endpgm
+;
+; GFX1264-LABEL: uniform_fadd_v2bf16:
+; GFX1264:       ; %bb.0:
+; GFX1264-NEXT:    s_clause 0x1
+; GFX1264-NEXT:    s_load_b128 s[8:11], s[4:5], 0x24
+; GFX1264-NEXT:    s_load_b32 s0, s[4:5], 0x34
+; GFX1264-NEXT:    s_mov_b64 s[2:3], 0
+; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    s_load_b32 s1, s[10:11], 0x0
+; GFX1264-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX1264-NEXT:    s_and_b32 s13, s0, 0xffff0000
+; GFX1264-NEXT:    s_mov_b32 s4, s10
+; GFX1264-NEXT:    s_mov_b32 s5, s11
+; GFX1264-NEXT:    s_wait_kmcnt 0x0
+; GFX1264-NEXT:    v_mov_b32_e32 v1, s1
+; GFX1264-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1264-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1264-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1264-NEXT:    v_add_f32_e32 v0, s12, v0
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_add_f32_e32 v2, s13, v2
+; GFX1264-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1264-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1264-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1264-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1264-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; GFX1264-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1264-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1264-NEXT:    v_cmp_u_f32_e64 s[0:1], v0, v0
+; GFX1264-NEXT:    s_wait_alu 0xfffd
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc
+; GFX1264-NEXT:    s_wait_alu 0xf1ff
+; GFX1264-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s[0:1]
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1264-NEXT:    v_mov_b32_e32 v3, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v2, v0
+; GFX1264-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1264-NEXT:    s_wait_loadcnt 0x0
+; GFX1264-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1264-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1264-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1264-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1264-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX1264-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1264-NEXT:    s_mov_b32 s10, -1
+; GFX1264-NEXT:    buffer_store_b32 v2, off, s[8:11], null
+; GFX1264-NEXT:    s_endpgm
+;
+; GFX1232-LABEL: uniform_fadd_v2bf16:
+; GFX1232:       ; %bb.0:
+; GFX1232-NEXT:    s_clause 0x1
+; GFX1232-NEXT:    s_load_b128 s[8:11], s[4:5], 0x24
+; GFX1232-NEXT:    s_load_b32 s0, s[4:5], 0x34
+; GFX1232-NEXT:    s_mov_b32 s1, 0
+; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    s_load_b32 s4, s[10:11], 0x0
+; GFX1232-NEXT:    s_lshl_b32 s2, s0, 16
+; GFX1232-NEXT:    s_and_b32 s3, s0, 0xffff0000
+; GFX1232-NEXT:    s_mov_b32 s5, s11
+; GFX1232-NEXT:    s_wait_kmcnt 0x0
+; GFX1232-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1232-NEXT:    s_mov_b32 s4, s10
+; GFX1232-NEXT:  .LBB21_1: ; %atomicrmw.start
+; GFX1232-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1232-NEXT:    v_lshlrev_b32_e32 v0, 16, v1
+; GFX1232-NEXT:    v_and_b32_e32 v2, 0xffff0000, v1
+; GFX1232-NEXT:    v_add_f32_e32 v0, s2, v0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1232-NEXT:    v_add_f32_e32 v2, s3, v2
+; GFX1232-NEXT:    v_bfe_u32 v3, v0, 16, 1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1232-NEXT:    v_bfe_u32 v4, v2, 16, 1
+; GFX1232-NEXT:    v_or_b32_e32 v5, 0x400000, v0
+; GFX1232-NEXT:    v_or_b32_e32 v6, 0x400000, v2
+; GFX1232-NEXT:    v_cmp_u_f32_e32 vcc_lo, v2, v2
+; GFX1232-NEXT:    v_add3_u32 v3, v3, v0, 0x7fff
+; GFX1232-NEXT:    v_add3_u32 v4, v4, v2, 0x7fff
+; GFX1232-NEXT:    v_cmp_u_f32_e64 s0, v0, v0
+; GFX1232-NEXT:    s_wait_alu 0xfffd
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1232-NEXT:    v_cndmask_b32_e32 v2, v4, v6, vcc_lo
+; GFX1232-NEXT:    s_wait_alu 0xf1ff
+; GFX1232-NEXT:    v_cndmask_b32_e64 v0, v3, v5, s0
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
+; GFX1232-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1232-NEXT:    buffer_atomic_cmpswap_b32 v[2:3], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1232-NEXT:    s_wait_loadcnt 0x0
+; GFX1232-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v1
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v2
+; GFX1232-NEXT:    s_or_b32 s1, vcc_lo, s1
+; GFX1232-NEXT:    s_wait_alu 0xfffe
+; GFX1232-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
+; GFX1232-NEXT:    s_cbranch_execnz .LBB21_1
+; GFX1232-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s1
+; GFX1232-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX1232-NEXT:    s_mov_b32 s10, -1
+; GFX1232-NEXT:    buffer_store_b32 v2, off, s[8:11], null
+; GFX1232-NEXT:    s_endpgm
+  %rmw = atomicrmw fadd ptr addrspace(1) %uniform.ptr, <2 x bfloat> %val monotonic, align 4
+  store <2 x bfloat> %rmw, ptr addrspace(1) %result
+  ret void
+}


        


More information about the llvm-commits mailing list