Merging r360293:

------------------------------------------------------------------------
r360293 | arsenm | 2019-05-08 15:09:57 -0700 (Wed, 08 May 2019) | 21 lines

AMDGPU: Select VOP3 form of add

The VOP3 form should always be the preferred selection, to be shrunk
later. This should only be an optimization issue, but this partially
works around a problem from clobbering VCC when SIFixSGPRCopies
rewrites an SCC defining operation directly to VCC.

3 of the testcases are regressions from failing to fold the immediate
in cases it should. These can be avoided by improving the VCC liveness
handling in SIFoldOperands. Simply increasing the threshold to
computeRegisterLiveness works, although this is common enough that VCC
liveness should probably be tracked throughout the pass. The hack of
leaving behind an implicit_def instruction to avoid breaking iterator
wastes instruction count, which inhibits finding the VCC def in long
chains of adds. Doing this however exposes different, worse looking
regressions from poor scheduling behavior. This could probably be
avoided around by forcing the shrink of the addc here, but the
scheduler should probably be fixed.

The r600 add test needs to be split out because it asserts on the
arguments in the new test during the calling convention lowering.
------------------------------------------------------------------------

llvm-svn: 362658

8 files changed, 202 insertions, 74 deletions

diff --git a/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
index 77a042c..8cf524a 100644
--- a/llvm/lib/Target/AMDGPU/VOP2Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
@@ -515,17 +515,13 @@ let AddedComplexity = 1 in {
 }
 
 let SubtargetPredicate = HasAddNoCarryInsts in {
-  def : DivergentBinOp<add, V_ADD_U32_e32>;
-
+  def : DivergentBinOp<add, V_ADD_U32_e64>;
   def : DivergentBinOp<sub, V_SUB_U32_e64>;
 }
 
-
-def : DivergentBinOp<add, V_ADD_I32_e32>;
+def : DivergentBinOp<add, V_ADD_I32_e64>;
 def : DivergentBinOp<sub, V_SUB_I32_e64>;
 
-def : DivergentBinOp<sub, V_SUBREV_I32_e32>;
-
 def : DivergentBinOp<srl, V_LSHRREV_B32_e32>;
 def : DivergentBinOp<sra, V_ASHRREV_I32_e32>;
 def : DivergentBinOp<shl, V_LSHLREV_B32_e32>;
diff --git a/llvm/test/CodeGen/AMDGPU/add.ll b/llvm/test/CodeGen/AMDGPU/add.ll
index 6a108db8..bce924e 100644
--- a/llvm/test/CodeGen/AMDGPU/add.ll
+++ b/llvm/test/CodeGen/AMDGPU/add.ll
@@ -1,11 +1,8 @@
 ; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SIVI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SIVI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,FUNC %s
-; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -enable-var-scope -check-prefix=EG -check-prefix=FUNC %s
 
 ; FUNC-LABEL: {{^}}s_add_i32:
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
 ; GCN: s_add_i32 s[[REG:[0-9]+]], {{s[0-9]+, s[0-9]+}}
 ; GCN: v_mov_b32_e32 v[[V_REG:[0-9]+]], s[[REG]]
 ; GCN: buffer_store_dword v[[V_REG]],
@@ -19,9 +16,6 @@ define amdgpu_kernel void @s_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %
 }
 
 ; FUNC-LABEL: {{^}}s_add_v2i32:
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
 ; GCN: s_add_i32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
 ; GCN: s_add_i32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
 define amdgpu_kernel void @s_add_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
@@ -34,11 +28,6 @@ define amdgpu_kernel void @s_add_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> a
 }
 
 ; FUNC-LABEL: {{^}}s_add_v4i32:
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
 ; GCN: s_add_i32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
 ; GCN: s_add_i32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
 ; GCN: s_add_i32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
@@ -53,15 +42,6 @@ define amdgpu_kernel void @s_add_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> a
 }
 
 ; FUNC-LABEL: {{^}}s_add_v8i32:
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-
 ; GCN: s_add_i32
 ; GCN: s_add_i32
 ; GCN: s_add_i32
@@ -78,23 +58,6 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}s_add_v16i32:
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-; EG: ADD_INT
-
 ; GCN: s_add_i32
 ; GCN: s_add_i32
 ; GCN: s_add_i32
@@ -124,7 +87,7 @@ entry:
 ; SIVI: v_add_{{i|u}}32_e32 v{{[0-9]+}}, vcc, [[A]], [[B]]
 ; GFX9: v_add_u32_e32 v{{[0-9]+}}, [[A]], [[B]]
 define amdgpu_kernel void @v_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
-  %tid = call i32 @llvm.r600.read.tidig.x()
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
   %b_ptr = getelementptr i32, i32 addrspace(1)* %gep, i32 1
   %a = load volatile i32, i32 addrspace(1)* %gep
@@ -139,7 +102,7 @@ define amdgpu_kernel void @v_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %
 ; SIVI: v_add_{{i|u}}32_e32 v{{[0-9]+}}, vcc, 0x7b, [[A]]
 ; GFX9: v_add_u32_e32 v{{[0-9]+}}, 0x7b, [[A]]
 define amdgpu_kernel void @v_add_imm_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
-  %tid = call i32 @llvm.r600.read.tidig.x()
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
   %b_ptr = getelementptr i32, i32 addrspace(1)* %gep, i32 1
   %a = load volatile i32, i32 addrspace(1)* %gep
@@ -151,13 +114,6 @@ define amdgpu_kernel void @v_add_imm_i32(i32 addrspace(1)* %out, i32 addrspace(1
 ; FUNC-LABEL: {{^}}add64:
 ; GCN: s_add_u32
 ; GCN: s_addc_u32
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-DAG: ADDC_UINT
-; EG-DAG: ADD_INT
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-NOT: SUB
 define amdgpu_kernel void @add64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
 entry:
   %add = add i64 %a, %b
@@ -172,13 +128,6 @@ entry:
 
 ; FUNC-LABEL: {{^}}add64_sgpr_vgpr:
 ; GCN-NOT: v_addc_u32_e32 s
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-DAG: ADDC_UINT
-; EG-DAG: ADD_INT
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-NOT: SUB
 define amdgpu_kernel void @add64_sgpr_vgpr(i64 addrspace(1)* %out, i64 %a, i64 addrspace(1)* %in) {
 entry:
   %0 = load i64, i64 addrspace(1)* %in
@@ -191,13 +140,6 @@ entry:
 ; FUNC-LABEL: {{^}}add64_in_branch:
 ; GCN: s_add_u32
 ; GCN: s_addc_u32
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-DAG: ADDC_UINT
-; EG-DAG: ADD_INT
-; EG-DAG: ADD_INT {{[* ]*}}
-; EG-NOT: SUB
 define amdgpu_kernel void @add64_in_branch(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %a, i64 %b, i64 %c) {
 entry:
   %0 = icmp eq i64 %a, 0
@@ -217,7 +159,26 @@ endif:
   ret void
 }
 
-declare i32 @llvm.r600.read.tidig.x() #1
+; Make sure the VOP3 form of add is initially selected. Otherwise pair
+; of opies from/to VCC would be necessary
+
+; GCN-LABEL: {{^}}add_select_vop3:
+; SI: v_add_i32_e64 v0, s[0:1], s0, v0
+; VI: v_add_u32_e64 v0, s[0:1], s0, v0
+; GFX9: v_add_u32_e32 v0, s0, v0
+
+; GCN: ; def vcc
+; GCN: ds_write_b32
+; GCN: ; use vcc
+define amdgpu_ps void @add_select_vop3(i32 inreg %s, i32 %v) {
+  %vcc = call i64 asm sideeffect "; def vcc", "={vcc}"()
+  %sub = add i32 %v, %s
+  store i32 %sub, i32 addrspace(3)* undef
+  call void asm sideeffect "; use vcc", "{vcc}"(i64 %vcc)
+  ret void
+}
+
+declare i32 @llvm.amdgcn.workitem.id.x() #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone speculatable }
diff --git a/llvm/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll b/llvm/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll
index 5997e27..e2c7f1c 100644
--- a/llvm/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll
+++ b/llvm/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll
@@ -7,6 +7,8 @@ declare void @llvm.amdgcn.s.barrier() #1
 
 ; Function Attrs: nounwind
 ; CHECK-LABEL: {{^}}signed_ds_offset_addressing_loop:
+; SI: s_movk_i32 [[K_0X88:s[0-9]+]], 0x
+; SI: s_movk_i32 [[K_0X100:s[0-9]+]], 0x100
 ; CHECK: BB0_1:
 ; CHECK: v_add_i32_e32 [[VADDR:v[0-9]+]],
 ; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[VADDR]]
@@ -14,9 +16,9 @@ declare void @llvm.amdgcn.s.barrier() #1
 ; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[VADDR8]]
 ; SI-DAG: v_add_i32_e32 [[VADDR0x80:v[0-9]+]], vcc, 0x80, [[VADDR]]
 ; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[VADDR0x80]]
-; SI-DAG: v_add_i32_e32 [[VADDR0x88:v[0-9]+]], vcc, 0x88, [[VADDR]]
+; SI-DAG: v_add_i32_e32 [[VADDR0x88:v[0-9]+]], vcc, [[K_0X88]], [[VADDR]]
 ; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[VADDR0x88]]
-; SI-DAG: v_add_i32_e32 [[VADDR0x100:v[0-9]+]], vcc, 0x100, [[VADDR]]
+; SI-DAG: v_add_i32_e32 [[VADDR0x100:v[0-9]+]], vcc, [[K_0X100]], [[VADDR]]
 ; SI-DAG: ds_read_b32 v{{[0-9]+}}, [[VADDR0x100]]
 
 ; CI-DAG: ds_read2_b32 v{{\[[0-9]+:[0-9]+\]}}, [[VADDR]] offset1:2
diff --git a/llvm/test/CodeGen/AMDGPU/fence-barrier.ll b/llvm/test/CodeGen/AMDGPU/fence-barrier.ll
index 8f5a06d..7de4f179 100644
--- a/llvm/test/CodeGen/AMDGPU/fence-barrier.ll
+++ b/llvm/test/CodeGen/AMDGPU/fence-barrier.ll
@@ -54,7 +54,8 @@ define amdgpu_kernel void @test_local(i32 addrspace(1)*) {
 }
 
 ; GCN-LABEL: {{^}}test_global
-; GCN: v_add_u32_e32 v{{[0-9]+}}, vcc, 0x888, v{{[0-9]+}}
+; GCN: s_movk_i32 [[K:s[0-9]+]], 0x888
+; GCN: v_add_u32_e32 v{{[0-9]+}}, vcc, [[K]], v{{[0-9]+}}
 ; GCN: flat_store_dword
 ; GCN: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
 ; GCN-NEXT: s_barrier
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
index 18abf60..77d518f 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
@@ -15,7 +15,8 @@ define amdgpu_kernel void @dpp_test(i32 addrspace(1)* %out, i32 %in1, i32 %in2)
 }
 
 ; VI-LABEL: {{^}}dpp_test1:
-; VI: v_add_u32_e32 [[REG:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; VI-OPT: v_add_u32_e32 [[REG:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; VI-NOOPT: v_add_u32_e64 [[REG:v[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; VI-NOOPT: v_mov_b32_e32 v{{[0-9]+}}, 0
 ; VI-NEXT: s_nop 0
 ; VI-NEXT: s_nop 0
diff --git a/llvm/test/CodeGen/AMDGPU/r600.add.ll b/llvm/test/CodeGen/AMDGPU/r600.add.ll
new file mode 100644
index 0000000..73eea3e
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/r600.add.ll
@@ -0,0 +1,167 @@
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -enable-var-scope -check-prefix=EG -check-prefix=FUNC %s
+
+; FUNC-LABEL: {{^}}s_add_i32:
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+define amdgpu_kernel void @s_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
+  %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
+  %a = load i32, i32 addrspace(1)* %in
+  %b = load i32, i32 addrspace(1)* %b_ptr
+  %result = add i32 %a, %b
+  store i32 %result, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}s_add_v2i32:
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+define amdgpu_kernel void @s_add_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
+  %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
+  %a = load <2 x i32>, <2 x i32> addrspace(1)* %in
+  %b = load <2 x i32>, <2 x i32> addrspace(1)* %b_ptr
+  %result = add <2 x i32> %a, %b
+  store <2 x i32> %result, <2 x i32> addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}s_add_v4i32:
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG: ADD_INT {{[* ]*}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+define amdgpu_kernel void @s_add_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
+  %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
+  %a = load <4 x i32>, <4 x i32> addrspace(1)* %in
+  %b = load <4 x i32>, <4 x i32> addrspace(1)* %b_ptr
+  %result = add <4 x i32> %a, %b
+  store <4 x i32> %result, <4 x i32> addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}s_add_v8i32:
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+define amdgpu_kernel void @s_add_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> %a, <8 x i32> %b) {
+entry:
+  %0 = add <8 x i32> %a, %b
+  store <8 x i32> %0, <8 x i32> addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}s_add_v16i32:
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+; EG: ADD_INT
+define amdgpu_kernel void @s_add_v16i32(<16 x i32> addrspace(1)* %out, <16 x i32> %a, <16 x i32> %b) {
+entry:
+  %0 = add <16 x i32> %a, %b
+  store <16 x i32> %0, <16 x i32> addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_add_i32:
+define amdgpu_kernel void @v_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
+  %tid = call i32 @llvm.r600.read.tidig.x()
+  %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
+  %b_ptr = getelementptr i32, i32 addrspace(1)* %gep, i32 1
+  %a = load volatile i32, i32 addrspace(1)* %gep
+  %b = load volatile i32, i32 addrspace(1)* %b_ptr
+  %result = add i32 %a, %b
+  store i32 %result, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_add_imm_i32:
+define amdgpu_kernel void @v_add_imm_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
+  %tid = call i32 @llvm.r600.read.tidig.x()
+  %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
+  %b_ptr = getelementptr i32, i32 addrspace(1)* %gep, i32 1
+  %a = load volatile i32, i32 addrspace(1)* %gep
+  %result = add i32 %a, 123
+  store i32 %result, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}add64:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-DAG: ADDC_UINT
+; EG-DAG: ADD_INT
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-NOT: SUB
+define amdgpu_kernel void @add64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
+entry:
+  %add = add i64 %a, %b
+  store i64 %add, i64 addrspace(1)* %out
+  ret void
+}
+
+; The v_addc_u32 and v_add_i32 instruction can't read SGPRs, because they
+; use VCC.  The test is designed so that %a will be stored in an SGPR and
+; %0 will be stored in a VGPR, so the comiler will be forced to copy %a
+; to a VGPR before doing the add.
+
+; FUNC-LABEL: {{^}}add64_sgpr_vgpr:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-DAG: ADDC_UINT
+; EG-DAG: ADD_INT
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-NOT: SUB
+define amdgpu_kernel void @add64_sgpr_vgpr(i64 addrspace(1)* %out, i64 %a, i64 addrspace(1)* %in) {
+entry:
+  %0 = load i64, i64 addrspace(1)* %in
+  %1 = add i64 %a, %0
+  store i64 %1, i64 addrspace(1)* %out
+  ret void
+}
+
+; Test i64 add inside a branch.
+; FUNC-LABEL: {{^}}add64_in_branch:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[LO:T[0-9]+\.XY]]
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-DAG: ADDC_UINT
+; EG-DAG: ADD_INT
+; EG-DAG: ADD_INT {{[* ]*}}
+; EG-NOT: SUB
+define amdgpu_kernel void @add64_in_branch(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %a, i64 %b, i64 %c) {
+entry:
+  %0 = icmp eq i64 %a, 0
+  br i1 %0, label %if, label %else
+
+if:
+  %1 = load i64, i64 addrspace(1)* %in
+  br label %endif
+
+else:
+  %2 = add i64 %a, %b
+  br label %endif
+
+endif:
+  %3 = phi i64 [%1, %if], [%2, %else]
+  store i64 %3, i64 addrspace(1)* %out
+  ret void
+}
+
+declare i32 @llvm.r600.read.tidig.x() #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readnone speculatable }
diff --git a/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll b/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
index 9b469621..cf42ee9 100644
--- a/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
+++ b/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
@@ -458,7 +458,7 @@ bb7:                                              ; preds = %bb3
 }
 
 ; GCN-LABEL: {{^}}phi_visit_order:
-; GCN: v_add_i32_e32 v{{[0-9]+}}, vcc, 1, v{{[0-9]+}}
+; GCN: v_add_i32_e64 v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, 1, v{{[0-9]+}}
 define amdgpu_kernel void @phi_visit_order() {
 bb:
   br label %bb1
diff --git a/llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll b/llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll
index 748f047..4a4203d 100644
--- a/llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll
+++ b/llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll
@@ -12,7 +12,7 @@
 ; LINE: v_mov_b32_e32 v{{[0-9]+}}, 0x888
 ; LINE: ; {{.*}}source-lines.cl:3
 ; LINE: ; {{.*}}source-lines.cl:4
-; LINE: v_add_u32_e32
+; LINE: v_add_u32_e64
 ; LINE: ; {{.*}}source-lines.cl:5
 ; LINE: flat_store_dword
 ; Epilogue.
@@ -28,7 +28,7 @@
 ; SOURCE: v_mov_b32_e32 v{{[0-9]+}}, 0x888
 ; SOURCE: ; int var1 = 0x888;
 ; SOURCE: ; int var2 = var0 + var1;
-; SOURCE: v_add_u32_e32
+; SOURCE: v_add_u32_e64
 ; SOURCE: ; *Out = var2;
 ; SOURCE: flat_store_dword
 ; Epilogue.