-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG-CHECK %s
-; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck --check-prefix=CM-CHECK %s
-; RUN: llc < %s -march=r600 -mcpu=verde | FileCheck --check-prefix=SI-CHECK %s
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG-CHECK --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck --check-prefix=CM-CHECK --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -mcpu=verde -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK --check-prefix=FUNC %s
;===------------------------------------------------------------------------===;
; Global Address Space
;===------------------------------------------------------------------------===;
+; FUNC-LABEL: @store_i1
+; EG-CHECK: MEM_RAT MSKOR
+; SI-CHECK: BUFFER_STORE_BYTE
+define void @store_i1(i1 addrspace(1)* %out) {
+entry:
+ store i1 true, i1 addrspace(1)* %out
+ ret void
+}
; i8 store
-; EG-CHECK: @store_i8
+; EG-CHECK-LABEL: @store_i8
; EG-CHECK: MEM_RAT MSKOR T[[RW_GPR:[0-9]]].XW, T{{[0-9]}}.X
; EG-CHECK: VTX_READ_8 [[VAL:T[0-9]\.X]], [[VAL]]
; IG 0: Get the byte index and truncate the value
; EG-CHECK: MOV T[[RW_GPR]].Y, 0.0
; EG-CHECK: MOV * T[[RW_GPR]].Z, 0.0
-; SI-CHECK: @store_i8
+; SI-CHECK-LABEL: @store_i8
; SI-CHECK: BUFFER_STORE_BYTE
define void @store_i8(i8 addrspace(1)* %out, i8 %in) {
}
; i16 store
-; EG-CHECK: @store_i16
+; EG-CHECK-LABEL: @store_i16
; EG-CHECK: MEM_RAT MSKOR T[[RW_GPR:[0-9]]].XW, T{{[0-9]}}.X
; EG-CHECK: VTX_READ_16 [[VAL:T[0-9]\.X]], [[VAL]]
; IG 0: Get the byte index and truncate the value
; EG-CHECK: MOV T[[RW_GPR]].Y, 0.0
; EG-CHECK: MOV * T[[RW_GPR]].Z, 0.0
-; SI-CHECK: @store_i16
+; SI-CHECK-LABEL: @store_i16
; SI-CHECK: BUFFER_STORE_SHORT
define void @store_i16(i16 addrspace(1)* %out, i16 %in) {
entry:
ret void
}
-; EG-CHECK: @store_v2i8
+; EG-CHECK-LABEL: @store_v2i8
; EG-CHECK: MEM_RAT MSKOR
; EG-CHECK-NOT: MEM_RAT MSKOR
-; SI-CHECK: @store_v2i8
+; SI-CHECK-LABEL: @store_v2i8
; SI-CHECK: BUFFER_STORE_BYTE
; SI-CHECK: BUFFER_STORE_BYTE
define void @store_v2i8(<2 x i8> addrspace(1)* %out, <2 x i32> %in) {
}
-; EG-CHECK: @store_v2i16
+; EG-CHECK-LABEL: @store_v2i16
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v2i16
+; CM-CHECK-LABEL: @store_v2i16
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v2i16
-; SI-CHECK: BUFFER_STORE_DWORD
+; SI-CHECK-LABEL: @store_v2i16
+; SI-CHECK: BUFFER_STORE_SHORT
+; SI-CHECK: BUFFER_STORE_SHORT
define void @store_v2i16(<2 x i16> addrspace(1)* %out, <2 x i32> %in) {
entry:
%0 = trunc <2 x i32> %in to <2 x i16>
ret void
}
-; EG-CHECK: @store_v4i8
+; EG-CHECK-LABEL: @store_v4i8
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v4i8
+; CM-CHECK-LABEL: @store_v4i8
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v4i8
+; SI-CHECK-LABEL: @store_v4i8
; SI-CHECK: BUFFER_STORE_BYTE
; SI-CHECK: BUFFER_STORE_BYTE
; SI-CHECK: BUFFER_STORE_BYTE
}
; floating-point store
-; EG-CHECK: @store_f32
+; EG-CHECK-LABEL: @store_f32
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+\.X, T[0-9]+\.X}}, 1
-; CM-CHECK: @store_f32
+; CM-CHECK-LABEL: @store_f32
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: @store_f32
+; SI-CHECK-LABEL: @store_f32
; SI-CHECK: BUFFER_STORE_DWORD
define void @store_f32(float addrspace(1)* %out, float %in) {
ret void
}
-; EG-CHECK: @store_v4i16
+; EG-CHECK-LABEL: @store_v4i16
; EG-CHECK: MEM_RAT MSKOR
; EG-CHECK: MEM_RAT MSKOR
; EG-CHECK: MEM_RAT MSKOR
; EG-CHECK: MEM_RAT MSKOR
; EG-CHECK-NOT: MEM_RAT MSKOR
-; SI-CHECK: @store_v4i16
+; SI-CHECK-LABEL: @store_v4i16
; SI-CHECK: BUFFER_STORE_SHORT
; SI-CHECK: BUFFER_STORE_SHORT
; SI-CHECK: BUFFER_STORE_SHORT
}
; vec2 floating-point stores
-; EG-CHECK: @store_v2f32
+; EG-CHECK-LABEL: @store_v2f32
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v2f32
+; CM-CHECK-LABEL: @store_v2f32
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v2f32
+; SI-CHECK-LABEL: @store_v2f32
; SI-CHECK: BUFFER_STORE_DWORDX2
define void @store_v2f32(<2 x float> addrspace(1)* %out, float %a, float %b) {
ret void
}
-; EG-CHECK: @store_v4i32
+; EG-CHECK-LABEL: @store_v4i32
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
; EG-CHECK-NOT: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v4i32
+; CM-CHECK-LABEL: @store_v4i32
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
; CM-CHECK-NOT: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v4i32
+; SI-CHECK-LABEL: @store_v4i32
; SI-CHECK: BUFFER_STORE_DWORDX4
define void @store_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %in) {
entry:
ret void
}
+; FUNC-LABEL: @store_i64_i8
+; EG-CHECK: MEM_RAT MSKOR
+; SI-CHECK: BUFFER_STORE_BYTE
+define void @store_i64_i8(i8 addrspace(1)* %out, i64 %in) {
+entry:
+ %0 = trunc i64 %in to i8
+ store i8 %0, i8 addrspace(1)* %out
+ ret void
+}
+
+; FUNC-LABEL: @store_i64_i16
+; EG-CHECK: MEM_RAT MSKOR
+; SI-CHECK: BUFFER_STORE_SHORT
+define void @store_i64_i16(i16 addrspace(1)* %out, i64 %in) {
+entry:
+ %0 = trunc i64 %in to i16
+ store i16 %0, i16 addrspace(1)* %out
+ ret void
+}
+
;===------------------------------------------------------------------------===;
; Local Address Space
;===------------------------------------------------------------------------===;
-; EG-CHECK: @store_local_i8
+; FUNC-LABEL: @store_local_i1
; EG-CHECK: LDS_BYTE_WRITE
-; SI-CHECK: @store_local_i8
+; SI-CHECK: DS_WRITE_B8
+define void @store_local_i1(i1 addrspace(3)* %out) {
+entry:
+ store i1 true, i1 addrspace(3)* %out
+ ret void
+}
+
+; EG-CHECK-LABEL: @store_local_i8
+; EG-CHECK: LDS_BYTE_WRITE
+; SI-CHECK-LABEL: @store_local_i8
; SI-CHECK: DS_WRITE_B8
define void @store_local_i8(i8 addrspace(3)* %out, i8 %in) {
store i8 %in, i8 addrspace(3)* %out
ret void
}
-; EG-CHECK: @store_local_i16
+; EG-CHECK-LABEL: @store_local_i16
; EG-CHECK: LDS_SHORT_WRITE
-; SI-CHECK: @store_local_i16
+; SI-CHECK-LABEL: @store_local_i16
; SI-CHECK: DS_WRITE_B16
define void @store_local_i16(i16 addrspace(3)* %out, i16 %in) {
store i16 %in, i16 addrspace(3)* %out
ret void
}
-; EG-CHECK: @store_local_v2i16
+; EG-CHECK-LABEL: @store_local_v2i16
; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v2i16
+; CM-CHECK-LABEL: @store_local_v2i16
; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v2i16
-; SI-CHECK: DS_WRITE_B32
+; SI-CHECK-LABEL: @store_local_v2i16
+; SI-CHECK: DS_WRITE_B16
+; SI-CHECK: DS_WRITE_B16
define void @store_local_v2i16(<2 x i16> addrspace(3)* %out, <2 x i16> %in) {
entry:
store <2 x i16> %in, <2 x i16> addrspace(3)* %out
ret void
}
-; EG-CHECK: @store_local_v4i8
+; EG-CHECK-LABEL: @store_local_v4i8
; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v4i8
+; CM-CHECK-LABEL: @store_local_v4i8
; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v4i8
+; SI-CHECK-LABEL: @store_local_v4i8
; SI-CHECK: DS_WRITE_B8
; SI-CHECK: DS_WRITE_B8
; SI-CHECK: DS_WRITE_B8
ret void
}
-; EG-CHECK: @store_local_v2i32
+; EG-CHECK-LABEL: @store_local_v2i32
; EG-CHECK: LDS_WRITE
; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v2i32
+; CM-CHECK-LABEL: @store_local_v2i32
; CM-CHECK: LDS_WRITE
; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v2i32
-; SI-CHECK: DS_WRITE_B32
-; SI-CHECK: DS_WRITE_B32
+; SI-CHECK-LABEL: @store_local_v2i32
+; SI-CHECK: DS_WRITE_B64
define void @store_local_v2i32(<2 x i32> addrspace(3)* %out, <2 x i32> %in) {
entry:
store <2 x i32> %in, <2 x i32> addrspace(3)* %out
ret void
}
-; EG-CHECK: @store_local_v4i32
+; EG-CHECK-LABEL: @store_local_v4i32
; EG-CHECK: LDS_WRITE
; EG-CHECK: LDS_WRITE
; EG-CHECK: LDS_WRITE
; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v4i32
+; CM-CHECK-LABEL: @store_local_v4i32
; CM-CHECK: LDS_WRITE
; CM-CHECK: LDS_WRITE
; CM-CHECK: LDS_WRITE
; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v4i32
+; SI-CHECK-LABEL: @store_local_v4i32
; SI-CHECK: DS_WRITE_B32
; SI-CHECK: DS_WRITE_B32
; SI-CHECK: DS_WRITE_B32
ret void
}
+; FUNC-LABEL: @store_local_i64_i8
+; EG-CHECK: LDS_BYTE_WRITE
+; SI-CHECK: DS_WRITE_B8
+define void @store_local_i64_i8(i8 addrspace(3)* %out, i64 %in) {
+entry:
+ %0 = trunc i64 %in to i8
+ store i8 %0, i8 addrspace(3)* %out
+ ret void
+}
+
+; FUNC-LABEL: @store_local_i64_i16
+; EG-CHECK: LDS_SHORT_WRITE
+; SI-CHECK: DS_WRITE_B16
+define void @store_local_i64_i16(i16 addrspace(3)* %out, i64 %in) {
+entry:
+ %0 = trunc i64 %in to i16
+ store i16 %0, i16 addrspace(3)* %out
+ ret void
+}
+
; The stores in this function are combined by the optimizer to create a
; 64-bit store with 32-bit alignment. This is legal for SI and the legalizer
; should not try to split the 64-bit store back into 2 32-bit stores.
; Evergreen / Northern Islands don't support 64-bit stores yet, so there should
; be two 32-bit stores.
-; EG-CHECK: @vecload2
+; EG-CHECK-LABEL: @vecload2
; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @vecload2
+; CM-CHECK-LABEL: @vecload2
; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @vecload2
+; SI-CHECK-LABEL: @vecload2
; SI-CHECK: BUFFER_STORE_DWORDX2
define void @vecload2(i32 addrspace(1)* nocapture %out, i32 addrspace(2)* nocapture %mem) #0 {
entry:
- %0 = load i32 addrspace(2)* %mem, align 4, !tbaa !5
+ %0 = load i32 addrspace(2)* %mem, align 4
%arrayidx1.i = getelementptr inbounds i32 addrspace(2)* %mem, i64 1
- %1 = load i32 addrspace(2)* %arrayidx1.i, align 4, !tbaa !5
- store i32 %0, i32 addrspace(1)* %out, align 4, !tbaa !5
+ %1 = load i32 addrspace(2)* %arrayidx1.i, align 4
+ store i32 %0, i32 addrspace(1)* %out, align 4
%arrayidx1 = getelementptr inbounds i32 addrspace(1)* %out, i64 1
- store i32 %1, i32 addrspace(1)* %arrayidx1, align 4, !tbaa !5
+ store i32 %1, i32 addrspace(1)* %arrayidx1, align 4
ret void
}
attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
-!5 = metadata !{metadata !"int", metadata !6}
-!6 = metadata !{metadata !"omnipotent char", metadata !7}
-!7 = metadata !{metadata !"Simple C/C++ TBAA"}
+; When i128 was a legal type this program generated cannot select errors:
+
+; FUNC-LABEL: @i128-const-store
+; FIXME: We should be able to to this with one store instruction
+; EG-CHECK: STORE_RAW
+; EG-CHECK: STORE_RAW
+; EG-CHECK: STORE_RAW
+; EG-CHECK: STORE_RAW
+; CM-CHECK: STORE_DWORD
+; CM-CHECK: STORE_DWORD
+; CM-CHECK: STORE_DWORD
+; CM-CHECK: STORE_DWORD
+; SI: BUFFER_STORE_DWORDX2
+; SI: BUFFER_STORE_DWORDX2
+define void @i128-const-store(i32 addrspace(1)* %out) {
+entry:
+ store i32 1, i32 addrspace(1)* %out, align 4
+ %arrayidx2 = getelementptr inbounds i32 addrspace(1)* %out, i64 1
+ store i32 1, i32 addrspace(1)* %arrayidx2, align 4
+ %arrayidx4 = getelementptr inbounds i32 addrspace(1)* %out, i64 2
+ store i32 2, i32 addrspace(1)* %arrayidx4, align 4
+ %arrayidx6 = getelementptr inbounds i32 addrspace(1)* %out, i64 3
+ store i32 2, i32 addrspace(1)* %arrayidx6, align 4
+ ret void
+}
projects / oota-llvm.git / blobdiff