-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK --check-prefix=FUNC
-; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK --check-prefix=FUNC
-
-; This test checks that uses and defs of the AR register happen in the same
-; instruction clause.
-
-; FUNC-LABEL: @mova_same_clause
-
-; R600-CHECK: MOVA_INT
-; R600-CHECK-NOT: ALU clause
-; R600-CHECK: 0 + AR.x
-; R600-CHECK: MOVA_INT
-; R600-CHECK-NOT: ALU clause
-; R600-CHECK: 0 + AR.x
-
-; SI-CHECK: V_READFIRSTLANE
-; SI-CHECK: V_MOVRELD
-; SI-CHECK: S_CBRANCH
-; SI-CHECK: V_READFIRSTLANE
-; SI-CHECK: V_MOVRELD
-; SI-CHECK: S_CBRANCH
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
+; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
+; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
+
+declare i32 @llvm.r600.read.tidig.x() nounwind readnone
+
+; FUNC-LABEL: {{^}}mova_same_clause:
+
+; R600: LDS_WRITE
+; R600: LDS_WRITE
+; R600: LDS_READ
+; R600: LDS_READ
+
+; SI-PROMOTE: ds_write_b32
+; SI-PROMOTE: ds_write_b32
+; SI-PROMOTE: ds_read_b32
+; SI-PROMOTE: ds_read_b32
+
+; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
+; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
entry:
%stack = alloca [5 x i32], align 4
; XXX: This generated code has unnecessary MOVs, we should be able to optimize
; this.
-; FUNC-LABEL: @multiple_structs
-; R600-CHECK-NOT: MOVA_INT
-; SI-CHECK-NOT: V_MOVREL
+; FUNC-LABEL: {{^}}multiple_structs:
+; R600-NOT: MOVA_INT
+; SI-NOT: v_movrel
+; SI-NOT: v_movrel
%struct.point = type { i32, i32 }
define void @multiple_structs(i32 addrspace(1)* %out) {
; loads and stores should be lowered to copies, so there shouldn't be any
; MOVA instructions.
-; FUNC-LABEL: @direct_loop
-; R600-CHECK-NOT: MOVA_INT
-; SI-CHECK-NOT: V_MOVREL
+; FUNC-LABEL: {{^}}direct_loop:
+; R600-NOT: MOVA_INT
+; SI-NOT: v_movrel
define void @direct_loop(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
entry:
ret void
}
-; FUNC-LABEL: @short_array
+; FUNC-LABEL: {{^}}short_array:
-; R600-CHECK: MOV {{\** *}}T{{[0-9]\.[XYZW]}}, literal
-; R600-CHECK: 65536
-; R600-CHECK: MOVA_INT
+; R600: MOVA_INT
-; SI-CHECK: V_MOV_B32_e32 v{{[0-9]}}, 65536
-; SI-CHECK: V_MOVRELS_B32_e32
+; SI-PROMOTE-DAG: buffer_store_short v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x68,0xe0
+; SI-PROMOTE-DAG: buffer_store_short v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:2 ; encoding: [0x02,0x10,0x68,0xe0
+; SI-PROMOTE: buffer_load_sshort v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
define void @short_array(i32 addrspace(1)* %out, i32 %index) {
entry:
%0 = alloca [2 x i16]
ret void
}
-; FUNC-LABEL: @char_array
+; FUNC-LABEL: {{^}}char_array:
-; R600-CHECK: OR_INT {{\** *}}T{{[0-9]\.[XYZW]}}, {{[PVT0-9]+\.[XYZW]}}, literal
-; R600-CHECK: 256
-; R600-CHECK: MOVA_INT
+; R600: MOVA_INT
-; SI-CHECK: V_OR_B32_e32 v{{[0-9]}}, 256
-; SI-CHECK: V_MOVRELS_B32_e32
+; SI-DAG: buffer_store_byte v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x60,0xe0
+; SI-DAG: buffer_store_byte v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:1 ; encoding: [0x01,0x10,0x60,0xe0
define void @char_array(i32 addrspace(1)* %out, i32 %index) {
entry:
%0 = alloca [2 x i8]
ret void
}
+
+; Make sure we don't overwrite workitem information with private memory
+
+; FUNC-LABEL: {{^}}work_item_info:
+; R600-NOT: MOV T0.X
+; Additional check in case the move ends up in the last slot
+; R600-NOT: MOV * TO.X
+
+; SI-NOT: v_mov_b32_e{{(32|64)}} v0
+define void @work_item_info(i32 addrspace(1)* %out, i32 %in) {
+entry:
+ %0 = alloca [2 x i32]
+ %1 = getelementptr [2 x i32]* %0, i32 0, i32 0
+ %2 = getelementptr [2 x i32]* %0, i32 0, i32 1
+ store i32 0, i32* %1
+ store i32 1, i32* %2
+ %3 = getelementptr [2 x i32]* %0, i32 0, i32 %in
+ %4 = load i32* %3
+ %5 = call i32 @llvm.r600.read.tidig.x()
+ %6 = add i32 %4, %5
+ store i32 %6, i32 addrspace(1)* %out
+ ret void
+}
+
+; Test that two stack objects are not stored in the same register
+; The second stack object should be in T3.X
+; FUNC-LABEL: {{^}}no_overlap:
+; R600_CHECK: MOV
+; R600_CHECK: [[CHAN:[XYZW]]]+
+; R600-NOT: [[CHAN]]+
+; SI: v_mov_b32_e32 v3
+define void @no_overlap(i32 addrspace(1)* %out, i32 %in) {
+entry:
+ %0 = alloca [3 x i8], align 1
+ %1 = alloca [2 x i8], align 1
+ %2 = getelementptr [3 x i8]* %0, i32 0, i32 0
+ %3 = getelementptr [3 x i8]* %0, i32 0, i32 1
+ %4 = getelementptr [3 x i8]* %0, i32 0, i32 2
+ %5 = getelementptr [2 x i8]* %1, i32 0, i32 0
+ %6 = getelementptr [2 x i8]* %1, i32 0, i32 1
+ store i8 0, i8* %2
+ store i8 1, i8* %3
+ store i8 2, i8* %4
+ store i8 1, i8* %5
+ store i8 0, i8* %6
+ %7 = getelementptr [3 x i8]* %0, i32 0, i32 %in
+ %8 = getelementptr [2 x i8]* %1, i32 0, i32 %in
+ %9 = load i8* %7
+ %10 = load i8* %8
+ %11 = add i8 %9, %10
+ %12 = sext i8 %11 to i32
+ store i32 %12, i32 addrspace(1)* %out
+ ret void
+}
+
+define void @char_array_array(i32 addrspace(1)* %out, i32 %index) {
+entry:
+ %alloca = alloca [2 x [2 x i8]]
+ %gep0 = getelementptr [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 0
+ %gep1 = getelementptr [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 1
+ store i8 0, i8* %gep0
+ store i8 1, i8* %gep1
+ %gep2 = getelementptr [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 %index
+ %load = load i8* %gep2
+ %sext = sext i8 %load to i32
+ store i32 %sext, i32 addrspace(1)* %out
+ ret void
+}
+
+define void @i32_array_array(i32 addrspace(1)* %out, i32 %index) {
+entry:
+ %alloca = alloca [2 x [2 x i32]]
+ %gep0 = getelementptr [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 0
+ %gep1 = getelementptr [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 1
+ store i32 0, i32* %gep0
+ store i32 1, i32* %gep1
+ %gep2 = getelementptr [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 %index
+ %load = load i32* %gep2
+ store i32 %load, i32 addrspace(1)* %out
+ ret void
+}
+
+define void @i64_array_array(i64 addrspace(1)* %out, i32 %index) {
+entry:
+ %alloca = alloca [2 x [2 x i64]]
+ %gep0 = getelementptr [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 0
+ %gep1 = getelementptr [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 1
+ store i64 0, i64* %gep0
+ store i64 1, i64* %gep1
+ %gep2 = getelementptr [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 %index
+ %load = load i64* %gep2
+ store i64 %load, i64 addrspace(1)* %out
+ ret void
+}
+
+%struct.pair32 = type { i32, i32 }
+
+define void @struct_array_array(i32 addrspace(1)* %out, i32 %index) {
+entry:
+ %alloca = alloca [2 x [2 x %struct.pair32]]
+ %gep0 = getelementptr [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 0, i32 1
+ %gep1 = getelementptr [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 1, i32 1
+ store i32 0, i32* %gep0
+ store i32 1, i32* %gep1
+ %gep2 = getelementptr [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 %index, i32 0
+ %load = load i32* %gep2
+ store i32 %load, i32 addrspace(1)* %out
+ ret void
+}
+
+define void @struct_pair32_array(i32 addrspace(1)* %out, i32 %index) {
+entry:
+ %alloca = alloca [2 x %struct.pair32]
+ %gep0 = getelementptr [2 x %struct.pair32]* %alloca, i32 0, i32 0, i32 1
+ %gep1 = getelementptr [2 x %struct.pair32]* %alloca, i32 0, i32 1, i32 0
+ store i32 0, i32* %gep0
+ store i32 1, i32* %gep1
+ %gep2 = getelementptr [2 x %struct.pair32]* %alloca, i32 0, i32 %index, i32 0
+ %load = load i32* %gep2
+ store i32 %load, i32 addrspace(1)* %out
+ ret void
+}
+
+define void @select_private(i32 addrspace(1)* %out, i32 %in) nounwind {
+entry:
+ %tmp = alloca [2 x i32]
+ %tmp1 = getelementptr [2 x i32]* %tmp, i32 0, i32 0
+ %tmp2 = getelementptr [2 x i32]* %tmp, i32 0, i32 1
+ store i32 0, i32* %tmp1
+ store i32 1, i32* %tmp2
+ %cmp = icmp eq i32 %in, 0
+ %sel = select i1 %cmp, i32* %tmp1, i32* %tmp2
+ %load = load i32* %sel
+ store i32 %load, i32 addrspace(1)* %out
+ ret void
+}
+
+; AMDGPUPromoteAlloca does not know how to handle ptrtoint. When it
+; finds one, it should stop trying to promote.
+
+; FUNC-LABEL: ptrtoint:
+; SI-NOT: ds_write
+; SI: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen
+; SI: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:5
+define void @ptrtoint(i32 addrspace(1)* %out, i32 %a, i32 %b) {
+ %alloca = alloca [16 x i32]
+ %tmp0 = getelementptr [16 x i32]* %alloca, i32 0, i32 %a
+ store i32 5, i32* %tmp0
+ %tmp1 = ptrtoint [16 x i32]* %alloca to i32
+ %tmp2 = add i32 %tmp1, 5
+ %tmp3 = inttoptr i32 %tmp2 to i32*
+ %tmp4 = getelementptr i32* %tmp3, i32 %b
+ %tmp5 = load i32* %tmp4
+ store i32 %tmp5, i32 addrspace(1)* %out
+ ret void
+}