R600: Add missing tests for some intrinsics

llvm-svn: 212870

7 files changed, 109 insertions, 5 deletions

diff --git a/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.global.ll b/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.global.ll
new file mode 100644
index 00000000000..47f5255e501
--- /dev/null
+++ b/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.global.ll
@@ -0,0 +1,28 @@
+; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
+
+; FUNC-LABEL: @test_barrier_global
+; EG: GROUP_BARRIER
+; SI: S_BARRIER
+
+define void @test_barrier_global(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.x()
+  %1 = getelementptr i32 addrspace(1)* %out, i32 %0
+  store i32 %0, i32 addrspace(1)* %1
+  call void @llvm.AMDGPU.barrier.global()
+  %2 = call i32 @llvm.r600.read.local.size.x()
+  %3 = sub i32 %2, 1
+  %4 = sub i32 %3, %0
+  %5 = getelementptr i32 addrspace(1)* %out, i32 %4
+  %6 = load i32 addrspace(1)* %5
+  store i32 %6, i32 addrspace(1)* %1
+  ret void
+}
+
+declare void @llvm.AMDGPU.barrier.global()
+
+declare i32 @llvm.r600.read.tidig.x() #0
+declare i32 @llvm.r600.read.local.size.x() #0
+
+attributes #0 = { readnone }
diff --git a/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.local.ll b/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.local.ll
index 8d3c9ca2230..7203675bb47 100644
--- a/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.local.ll
+++ b/llvm/test/CodeGen/R600/llvm.AMDGPU.barrier.local.ll
@@ -1,8 +1,11 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
-; CHECK: GROUP_BARRIER
+; FUNC-LABEL: @test_barrier_local
+; EG: GROUP_BARRIER
+; SI: S_BARRIER
 
-define void @test(i32 addrspace(1)* %out) {
+define void @test_barrier_local(i32 addrspace(1)* %out) {
 entry:
   %0 = call i32 @llvm.r600.read.tidig.x()
   %1 = getelementptr i32 addrspace(1)* %out, i32 %0
@@ -17,8 +20,9 @@ entry:
   ret void
 }
 
-declare i32 @llvm.r600.read.tidig.x() #0
 declare void @llvm.AMDGPU.barrier.local()
+
+declare i32 @llvm.r600.read.tidig.x() #0
 declare i32 @llvm.r600.read.local.size.x() #0
 
 attributes #0 = { readnone }
diff --git a/llvm/test/CodeGen/R600/llvm.AMDGPU.kill.ll b/llvm/test/CodeGen/R600/llvm.AMDGPU.kill.ll
index 4ab6a8ae09f..1f82ffb53f1 100644
--- a/llvm/test/CodeGen/R600/llvm.AMDGPU.kill.ll
+++ b/llvm/test/CodeGen/R600/llvm.AMDGPU.kill.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=r600 -mcpu=verde -verify-machineinstrs | FileCheck --check-prefix=SI %s
+; RUN: llc -march=r600 -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
 
 ; SI-LABEL: @kill_gs_const
 ; SI-NOT: V_CMPX_LE_F32
diff --git a/llvm/test/CodeGen/R600/llvm.AMDGPU.rcp.f64.ll b/llvm/test/CodeGen/R600/llvm.AMDGPU.rcp.f64.ll
new file mode 100644
index 00000000000..62a30a79a2d
--- /dev/null
+++ b/llvm/test/CodeGen/R600/llvm.AMDGPU.rcp.f64.ll
@@ -0,0 +1,29 @@
+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+
+declare double @llvm.AMDGPU.rcp.f64(double) nounwind readnone
+declare double @llvm.sqrt.f64(double) nounwind readnone
+
+; FUNC-LABEL: @rcp_f64
+; SI: V_RCP_F64_e32
+define void @rcp_f64(double addrspace(1)* %out, double %src) nounwind {
+  %rcp = call double @llvm.AMDGPU.rcp.f64(double %src) nounwind readnone
+  store double %rcp, double addrspace(1)* %out, align 8
+  ret void
+}
+
+; FUNC-LABEL: @rcp_pat_f64
+; SI: V_RCP_F64_e32
+define void @rcp_pat_f64(double addrspace(1)* %out, double %src) nounwind {
+  %rcp = fdiv double 1.0, %src
+  store double %rcp, double addrspace(1)* %out, align 8
+  ret void
+}
+
+; FUNC-LABEL: @rsq_rcp_pat_f64
+; SI: V_RSQ_F64_e32
+define void @rsq_rcp_pat_f64(double addrspace(1)* %out, double %src) nounwind {
+  %sqrt = call double @llvm.sqrt.f64(double %src) nounwind readnone
+  %rcp = call double @llvm.AMDGPU.rcp.f64(double %sqrt) nounwind readnone
+  store double %rcp, double addrspace(1)* %out, align 8
+  ret void
+}
diff --git a/llvm/test/CodeGen/R600/llvm.amdgpu.dp4.ll b/llvm/test/CodeGen/R600/llvm.amdgpu.dp4.ll
new file mode 100644
index 00000000000..812b6a40ee5
--- /dev/null
+++ b/llvm/test/CodeGen/R600/llvm.amdgpu.dp4.ll
@@ -0,0 +1,11 @@
+; RUN: llc -march=r600 -mcpu=redwood -verify-machineinstrs < %s
+
+declare float @llvm.AMDGPU.dp4(<4 x float>, <4 x float>) nounwind readnone
+
+define void @test_dp4(float addrspace(1)* %out, <4 x float> addrspace(1)* %a, <4 x float> addrspace(1)* %b) nounwind {
+  %src0 = load <4 x float> addrspace(1)* %a, align 16
+  %src1 = load <4 x float> addrspace(1)* %b, align 16
+  %dp4 = call float @llvm.AMDGPU.dp4(<4 x float> %src0, <4 x float> %src1) nounwind readnone
+  store float %dp4, float addrspace(1)* %out, align 4
+  ret void
+}
diff --git a/llvm/test/CodeGen/R600/llvm.amdgpu.kilp.ll b/llvm/test/CodeGen/R600/llvm.amdgpu.kilp.ll
new file mode 100644
index 00000000000..1b8b1bfd208
--- /dev/null
+++ b/llvm/test/CodeGen/R600/llvm.amdgpu.kilp.ll
@@ -0,0 +1,20 @@
+; RUN: llc -march=r600 -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
+
+; SI-LABEL: @kilp_gs_const
+; SI: S_MOV_B64 exec, 0
+define void @kilp_gs_const() #0 {
+main_body:
+  %0 = icmp ule i32 0, 3
+  %1 = select i1 %0, float 1.000000e+00, float -1.000000e+00
+  call void @llvm.AMDGPU.kilp(float %1)
+  %2 = icmp ule i32 3, 0
+  %3 = select i1 %2, float 1.000000e+00, float -1.000000e+00
+  call void @llvm.AMDGPU.kilp(float %3)
+  ret void
+}
+
+declare void @llvm.AMDGPU.kilp(float)
+
+attributes #0 = { "ShaderType"="2" }
+
+!0 = metadata !{metadata !"const", null, i32 1}
diff --git a/llvm/test/CodeGen/R600/llvm.amdgpu.lrp.ll b/llvm/test/CodeGen/R600/llvm.amdgpu.lrp.ll
new file mode 100644
index 00000000000..c493a016e33
--- /dev/null
+++ b/llvm/test/CodeGen/R600/llvm.amdgpu.lrp.ll
@@ -0,0 +1,12 @@
+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+
+declare float @llvm.AMDGPU.lrp(float, float, float) nounwind readnone
+
+; FUNC-LABEL: @test_lrp
+; SI: V_SUB_F32
+; SI: V_MAD_F32
+define void @test_lrp(float addrspace(1)* %out, float %src0, float %src1, float %src2) nounwind {
+  %mad = call float @llvm.AMDGPU.lrp(float %src0, float %src1, float %src2) nounwind readnone
+  store float %mad, float addrspace(1)* %out, align 4
+  ret void
+}