Reapply "R600: Add new intrinsic to read work dimensions"

[oota-llvm.git] / test / CodeGen / R600 / work-item-intrinsics.ll

diff --git a/test/CodeGen/R600/work-item-intrinsics.ll b/test/CodeGen/R600/work-item-intrinsics.ll
index a1337ae1f4d6972b54af4a165ed09c6f28d97ed8..cbefe2577fdaef119ca3bc9b3848e64fa246c053 100644 (file)
--- a/test/CodeGen/R600/work-item-intrinsics.ll
+++ b/test/CodeGen/R600/work-item-intrinsics.ll
@@ -128,6 +128,20 @@ entry:
   ret void
 }
 
   ret void
 }
 

+; FUNC-LABEL: {{^}}get_work_dim:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV [[VAL]], KC0[2].Z
+
+; SI: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0xb
+; SI: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; SI: BUFFER_STORE_DWORD [[VVAL]]
+define void @get_work_dim (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.AMDGPU.read.workdim() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+

 ; The tgid values are stored in sgprs offset by the number of user sgprs.
 ; Currently we always use exactly 2 user sgprs for the pointer to the
 ; kernel arguments, but this may change in the future.
 ; The tgid values are stored in sgprs offset by the number of user sgprs.
 ; Currently we always use exactly 2 user sgprs for the pointer to the
 ; kernel arguments, but this may change in the future.


@@ -209,4 +223,6 @@ declare i32 @llvm.r600.read.tidig.x() #0
 declare i32 @llvm.r600.read.tidig.y() #0
 declare i32 @llvm.r600.read.tidig.z() #0
 
 declare i32 @llvm.r600.read.tidig.y() #0
 declare i32 @llvm.r600.read.tidig.z() #0
 

+declare i32 @llvm.AMDGPU.read.workdim() #0
+

 attributes #0 = { readnone }
 attributes #0 = { readnone }