AMDGPU/SI: Add support for R_AMDGPU_GOTPCREL

Reviewers: rafael, ruiu, tony-tye, arsenm, kzhuravl

Subscribers: arsenm, llvm-commits, kzhuravl

Differential Revision: http://reviews.llvm.org/D21484

llvm-svn: 275268

5 files changed, 208 insertions, 23 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/global-variable-relocs.ll b/llvm/test/CodeGen/AMDGPU/global-variable-relocs.ll
new file mode 100644
index 00000000000..c39394a3527
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/global-variable-relocs.ll
@@ -0,0 +1,203 @@
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji < %s | FileCheck %s
+
+@private = private addrspace(1) global [256 x i32] zeroinitializer
+@internal = internal addrspace(1) global [256 x i32] zeroinitializer
+@available_externally = available_externally addrspace(1) global [256 x i32] zeroinitializer
+@linkonce = linkonce addrspace(1) global [256 x i32] zeroinitializer
+@weak= weak addrspace(1) global [256 x i32] zeroinitializer
+@common = common addrspace(1) global [256 x i32] zeroinitializer
+@extern_weak = extern_weak addrspace(1) global [256 x i32]
+@linkonce_odr = linkonce_odr addrspace(1) global [256 x i32] zeroinitializer
+@weak_odr = weak_odr addrspace(1) global [256 x i32] zeroinitializer
+@external = external addrspace(1) global [256 x i32]
+@external_w_init = addrspace(1) global [256 x i32] zeroinitializer
+
+; CHECK-LABEL: private_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], private+8
+; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[ADDR_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[ADDR_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @private_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @private, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: internal_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], internal+8
+; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[ADDR_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[ADDR_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @internal_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @internal, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: available_externally_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], available_externally@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @available_externally_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @available_externally, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: linkonce_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @linkonce_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @linkonce, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: weak_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @weak_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @weak, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: common_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], common@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @common_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @common, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: extern_weak_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], extern_weak@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @extern_weak_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @extern_weak, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: linkonce_odr_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce_odr@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @linkonce_odr_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @linkonce_odr, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: weak_odr_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak_odr@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @weak_odr_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @weak_odr, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: external_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @external_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @external, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: external_w_init_test:
+; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external_w_init@GOTPCREL+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
+; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
+; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
+; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
+; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
+; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
+define void @external_w_init_test(i32 addrspace(1)* %out) {
+  %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @external_w_init, i32 0, i32 1
+  %val = load i32, i32 addrspace(1)* %ptr
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK: .local private
+; CHECK: .local internal
+; CHECK: .weak linkonce
+; CHECK: .weak weak
+; CHECK: .weak linkonce_odr
+; CHECK: .weak weak_odr
+; CHECK-NOT: external{{$}}
+; CHECK: .globl external_w_init
diff --git a/llvm/test/CodeGen/AMDGPU/global-zero-initializer.ll b/llvm/test/CodeGen/AMDGPU/global-zero-initializer.ll
deleted file mode 100644
index 59c3960c46e..00000000000
--- a/llvm/test/CodeGen/AMDGPU/global-zero-initializer.ll
+++ /dev/null
@@ -1,18 +0,0 @@
-; RUN: llc -march=amdgcn -mcpu=SI < %s 2>&1 | FileCheck %s
-; RUN: llc -march=amdgcn -mcpu=tonga < %s 2>&1 | FileCheck %s
-
-; CHECK: {{^}}load_init_global_global:
-; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], global+4
-; CHECK: s_addc_u32 s5, s[[PC_HI]], 0
-; CHECK: buffer_load_dword v{{[0-9]+}}, off, s{{\[}}[[ADDR_LO]]:7], 0 offset:40
-; CHECK: global:
-; CHECK: .zero 1024
-@global = addrspace(1) global [256 x i32] zeroinitializer
-
-define void @load_init_global_global(i32 addrspace(1)* %out, i1 %p) {
- %gep = getelementptr [256 x i32], [256 x i32] addrspace(1)* @global, i32 0, i32 10
-  %ld = load i32, i32 addrspace(1)* %gep
-  store i32 %ld, i32 addrspace(1)* %out
-  ret void
-}
diff --git a/llvm/test/CodeGen/AMDGPU/hsa-func.ll b/llvm/test/CodeGen/AMDGPU/hsa-func.ll
index 4aa003265ba..28c8b5d73b0 100644
--- a/llvm/test/CodeGen/AMDGPU/hsa-func.ll
+++ b/llvm/test/CodeGen/AMDGPU/hsa-func.ll
@@ -18,7 +18,7 @@
 
 ; ELF: SHT_NOTE
 ; ELF: 0000: 04000000 08000000 01000000 414D4400
-; ELF: 0010: 02000000 00000000 04000000 1B000000
+; ELF: 0010: 02000000 01000000 04000000 1B000000
 
 ; ELF: 0020: 03000000 414D4400 04000700 07000000
 ; ELF: 0030: 00000000 00000000 414D4400 414D4447
@@ -30,7 +30,7 @@
 ; ELF: Type: Function (0x2)
 ; ELF: }
 
-; HSA: .hsa_code_object_version 2,0
+; HSA: .hsa_code_object_version 2,1
 ; HSA-CI: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"
 
diff --git a/llvm/test/CodeGen/AMDGPU/hsa-note-no-func.ll b/llvm/test/CodeGen/AMDGPU/hsa-note-no-func.ll
index 67d82e2731e..1b4a0f3090b 100644
--- a/llvm/test/CodeGen/AMDGPU/hsa-note-no-func.ll
+++ b/llvm/test/CodeGen/AMDGPU/hsa-note-no-func.ll
@@ -2,7 +2,7 @@
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo | FileCheck --check-prefix=HSA --check-prefix=HSA-VI %s
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=fiji | FileCheck --check-prefix=HSA --check-prefix=HSA-FIJI %s
 
-; HSA: .hsa_code_object_version 2,0
+; HSA: .hsa_code_object_version 2,1
 ; HSA-CI: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"
 ; HSA-FIJI: .hsa_code_object_isa 8,0,3,"AMD","AMDGPU"
diff --git a/llvm/test/CodeGen/AMDGPU/hsa.ll b/llvm/test/CodeGen/AMDGPU/hsa.ll
index 1f07242d29a..82d7da188ca 100644
--- a/llvm/test/CodeGen/AMDGPU/hsa.ll
+++ b/llvm/test/CodeGen/AMDGPU/hsa.ll
@@ -18,7 +18,7 @@
 
 ; ELF: SHT_NOTE
 ; ELF: 0000: 04000000 08000000 01000000 414D4400
-; ELF: 0010: 02000000 00000000 04000000 1B000000
+; ELF: 0010: 02000000 01000000 04000000 1B000000
 ; ELF: 0020: 03000000 414D4400 04000700 07000000
 ; ELF: 0030: 00000000 00000000 414D4400 414D4447
 ; ELF: 0040: 50550000
@@ -29,7 +29,7 @@
 ; ELF: Type: AMDGPU_HSA_KERNEL (0xA)
 ; ELF: }
 
-; HSA: .hsa_code_object_version 2,0
+; HSA: .hsa_code_object_version 2,1
 ; HSA-CI: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"