Lines Matching refs:src0

652 // CHECK: call i32 @llvm.amdgcn.mbcnt.lo(i32 %src0, i32 %src1)
653 kernel void test_mbcnt_lo(global uint* out, uint src0, uint src1) {
654   *out = __builtin_amdgcn_mbcnt_lo(src0, src1);
658 // CHECK: call i32 @llvm.amdgcn.mbcnt.hi(i32 %src0, i32 %src1)
659 kernel void test_mbcnt_hi(global uint* out, uint src0, uint src1) {
660   *out = __builtin_amdgcn_mbcnt_hi(src0, src1);
664 // CHECK: tail call i32 @llvm.fshr.i32(i32 %src0, i32 %src1, i32 %src2)
665 kernel void test_alignbit(global uint* out, uint src0, uint src1, uint src2) {
666   *out = __builtin_amdgcn_alignbit(src0, src1, src2);
670 // CHECK: tail call i32 @llvm.amdgcn.alignbyte(i32 %src0, i32 %src1, i32 %src2)
671 kernel void test_alignbyte(global uint* out, uint src0, uint src1, uint src2) {
672   *out = __builtin_amdgcn_alignbyte(src0, src1, src2);
676 // CHECK: tail call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 %src1, i32 %src2)
677 kernel void test_ubfe(global uint* out, uint src0, uint src1, uint src2) {
678   *out = __builtin_amdgcn_ubfe(src0, src1, src2);
682 // CHECK: tail call i32 @llvm.amdgcn.sbfe.i32(i32 %src0, i32 %src1, i32 %src2)
683 kernel void test_sbfe(global uint* out, uint src0, uint src1, uint src2) {
684   *out = __builtin_amdgcn_sbfe(src0, src1, src2);
688 // CHECK: tail call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %src0, float %src1)
689 kernel void test_cvt_pkrtz(global half2* out, float src0, float src1) {
690   *out = __builtin_amdgcn_cvt_pkrtz(src0, src1);
694 // CHECK: tail call <2 x i16> @llvm.amdgcn.cvt.pknorm.i16(float %src0, float %src1)
695 kernel void test_cvt_pknorm_i16(global short2* out, float src0, float src1) {
696   *out = __builtin_amdgcn_cvt_pknorm_i16(src0, src1);
700 // CHECK: tail call <2 x i16> @llvm.amdgcn.cvt.pknorm.u16(float %src0, float %src1)
701 kernel void test_cvt_pknorm_u16(global ushort2* out, float src0, float src1) {
702   *out = __builtin_amdgcn_cvt_pknorm_u16(src0, src1);
706 // CHECK: tail call <2 x i16> @llvm.amdgcn.cvt.pk.i16(i32 %src0, i32 %src1)
707 kernel void test_cvt_pk_i16(global short2* out, int src0, int src1) {
708   *out = __builtin_amdgcn_cvt_pk_i16(src0, src1);
712 // CHECK: tail call <2 x i16> @llvm.amdgcn.cvt.pk.u16(i32 %src0, i32 %src1)
713 kernel void test_cvt_pk_u16(global ushort2* out, uint src0, uint src1) {
714   *out = __builtin_amdgcn_cvt_pk_u16(src0, src1);
718 // CHECK: tail call i32 @llvm.amdgcn.cvt.pk.u8.f32(float %src0, i32 %src1, i32 %src2)
719 kernel void test_cvt_pk_u8_f32(global uint* out, float src0, uint src1, uint src2) {
720   *out = __builtin_amdgcn_cvt_pk_u8_f32(src0, src1, src2);
724 // CHECK: tail call i32 @llvm.amdgcn.sad.u8(i32 %src0, i32 %src1, i32 %src2)
725 kernel void test_sad_u8(global uint* out, uint src0, uint src1, uint src2) {
726   *out = __builtin_amdgcn_sad_u8(src0, src1, src2);
730 // CHECK: call i32 @llvm.amdgcn.msad.u8(i32 %src0, i32 %src1, i32 %src2)
731 kernel void test_msad_u8(global uint* out, uint src0, uint src1, uint src2) {
732   *out = __builtin_amdgcn_msad_u8(src0, src1, src2);
736 // CHECK: call i32 @llvm.amdgcn.sad.hi.u8(i32 %src0, i32 %src1, i32 %src2)
737 kernel void test_sad_hi_u8(global uint* out, uint src0, uint src1, uint src2) {
738   *out = __builtin_amdgcn_sad_hi_u8(src0, src1, src2);
742 // CHECK: call i32 @llvm.amdgcn.sad.u16(i32 %src0, i32 %src1, i32 %src2)
743 kernel void test_sad_u16(global uint* out, uint src0, uint src1, uint src2) {
744   *out = __builtin_amdgcn_sad_u16(src0, src1, src2);
748 // CHECK: call i64 @llvm.amdgcn.qsad.pk.u16.u8(i64 %src0, i32 %src1, i64 %src2)
749 kernel void test_qsad_pk_u16_u8(global ulong* out, ulong src0, uint src1, ulong src2) {
750   *out = __builtin_amdgcn_qsad_pk_u16_u8(src0, src1, src2);
754 // CHECK: call i64 @llvm.amdgcn.mqsad.pk.u16.u8(i64 %src0, i32 %src1, i64 %src2)
755 kernel void test_mqsad_pk_u16_u8(global ulong* out, ulong src0, uint src1, ulong src2) {
756   *out = __builtin_amdgcn_mqsad_pk_u16_u8(src0, src1, src2);
760 // CHECK: call <4 x i32> @llvm.amdgcn.mqsad.u32.u8(i64 %src0, i32 %src1, <4 x i32> %src2)
761 kernel void test_mqsad_u32_u8(global uint4* out, ulong src0, uint src1, uint4 src2) {
762   *out = __builtin_amdgcn_mqsad_u32_u8(src0, src1, src2);