R600/SI: Add global atomicrmw or

v2: Add separate offset/no-offset tests Signed-off-by: Aaron Watry <[email protected]> Reviewed-by: Matt Arsenault <[email protected]> llvm-svn: 220108
cachemeifyoucan · Oct 17, 2014 · 8a911e6 · 8a911e6
1 parent 58c9992
commit 8a911e6
Show file tree

Hide file tree

Showing 3 changed files with 84 additions and 1 deletion.
diff --git a/llvm/lib/Target/R600/AMDGPUInstructions.td b/llvm/lib/Target/R600/AMDGPUInstructions.td
@@ -390,6 +390,7 @@ def atomic_add_global : global_binary_atomic_op<atomic_load_add>;
 def atomic_and_global : global_binary_atomic_op<atomic_load_and>;
 def atomic_max_global : global_binary_atomic_op<atomic_load_max>;
 def atomic_min_global : global_binary_atomic_op<atomic_load_min>;
+def atomic_or_global : global_binary_atomic_op<atomic_load_or>;
 def atomic_sub_global : global_binary_atomic_op<atomic_load_sub>;
 def atomic_umax_global : global_binary_atomic_op<atomic_load_umax>;
 def atomic_umin_global : global_binary_atomic_op<atomic_load_umin>;

diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td
@@ -919,7 +919,9 @@ defm BUFFER_ATOMIC_UMAX : MUBUF_Atomic <
 defm BUFFER_ATOMIC_AND : MUBUF_Atomic <
   0x00000039, "BUFFER_ATOMIC_AND", VReg_32, i32, atomic_and_global
 >;
-//def BUFFER_ATOMIC_OR : MUBUF_ <0x0000003a, "BUFFER_ATOMIC_OR", []>;
+defm BUFFER_ATOMIC_OR : MUBUF_Atomic <
+  0x0000003a, "BUFFER_ATOMIC_OR", VReg_32, i32, atomic_or_global
+>;
 //def BUFFER_ATOMIC_XOR : MUBUF_ <0x0000003b, "BUFFER_ATOMIC_XOR", []>;
 //def BUFFER_ATOMIC_INC : MUBUF_ <0x0000003c, "BUFFER_ATOMIC_INC", []>;
 //def BUFFER_ATOMIC_DEC : MUBUF_ <0x0000003d, "BUFFER_ATOMIC_DEC", []>;

diff --git a/llvm/test/CodeGen/R600/global_atomics.ll b/llvm/test/CodeGen/R600/global_atomics.ll
@@ -559,3 +559,83 @@ entry:
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
+
+; FUNC-LABEL: {{^}}atomic_or_i32_offset:
+; SI: BUFFER_ATOMIC_OR v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:0x10{{$}}
+define void @atomic_or_i32_offset(i32 addrspace(1)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32 addrspace(1)* %out, i32 4
+  %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_ret_offset:
+; SI: BUFFER_ATOMIC_OR [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:0x10 glc {{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_or_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32 addrspace(1)* %out, i32 4
+  %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_addr64_offset:
+; SI: BUFFER_ATOMIC_OR v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:0x10{{$}}
+define void @atomic_or_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i32 addrspace(1)* %ptr, i32 4
+  %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_ret_addr64_offset:
+; SI: BUFFER_ATOMIC_OR [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:0x10 glc{{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_or_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i32 addrspace(1)* %ptr, i32 4
+  %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32:
+; SI: BUFFER_ATOMIC_OR v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_or_i32(i32 addrspace(1)* %out, i32 %in) {
+entry:
+  %0  = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_ret:
+; SI: BUFFER_ATOMIC_OR [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_or_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
+entry:
+  %0  = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_addr64:
+; SI: BUFFER_ATOMIC_OR v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+define void @atomic_or_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %0  = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_or_i32_ret_addr64:
+; SI: BUFFER_ATOMIC_OR [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_or_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %0  = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}