; CHECK: vpbroadcastb (%
define <16 x i8> @BB16(i8* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i8* %ptr, align 4
+ %q = load i8, i8* %ptr, align 4
%q0 = insertelement <16 x i8> undef, i8 %q, i32 0
%q1 = insertelement <16 x i8> %q0, i8 %q, i32 1
%q2 = insertelement <16 x i8> %q1, i8 %q, i32 2
; CHECK: vpbroadcastb (%
define <32 x i8> @BB32(i8* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i8* %ptr, align 4
+ %q = load i8, i8* %ptr, align 4
%q0 = insertelement <32 x i8> undef, i8 %q, i32 0
%q1 = insertelement <32 x i8> %q0, i8 %q, i32 1
%q2 = insertelement <32 x i8> %q1, i8 %q, i32 2
define <8 x i16> @W16(i16* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i16* %ptr, align 4
+ %q = load i16, i16* %ptr, align 4
%q0 = insertelement <8 x i16> undef, i16 %q, i32 0
%q1 = insertelement <8 x i16> %q0, i16 %q, i32 1
%q2 = insertelement <8 x i16> %q1, i16 %q, i32 2
; CHECK: vpbroadcastw (%
define <16 x i16> @WW16(i16* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i16* %ptr, align 4
+ %q = load i16, i16* %ptr, align 4
%q0 = insertelement <16 x i16> undef, i16 %q, i32 0
%q1 = insertelement <16 x i16> %q0, i16 %q, i32 1
%q2 = insertelement <16 x i16> %q1, i16 %q, i32 2
%qf = insertelement <16 x i16> %qe, i16 %q, i32 15
ret <16 x i16> %qf
}
-; CHECK: vpbroadcastd (%
+; CHECK: vbroadcastss (%
define <4 x i32> @D32(i32* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i32* %ptr, align 4
+ %q = load i32, i32* %ptr, align 4
%q0 = insertelement <4 x i32> undef, i32 %q, i32 0
%q1 = insertelement <4 x i32> %q0, i32 %q, i32 1
%q2 = insertelement <4 x i32> %q1, i32 %q, i32 2
%q3 = insertelement <4 x i32> %q2, i32 %q, i32 3
ret <4 x i32> %q3
}
-; CHECK: vpbroadcastd (%
+; CHECK: vbroadcastss (%
define <8 x i32> @DD32(i32* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i32* %ptr, align 4
+ %q = load i32, i32* %ptr, align 4
%q0 = insertelement <8 x i32> undef, i32 %q, i32 0
%q1 = insertelement <8 x i32> %q0, i32 %q, i32 1
%q2 = insertelement <8 x i32> %q1, i32 %q, i32 2
; CHECK: vpbroadcastq (%
define <2 x i64> @Q64(i64* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i64* %ptr, align 4
+ %q = load i64, i64* %ptr, align 4
%q0 = insertelement <2 x i64> undef, i64 %q, i32 0
%q1 = insertelement <2 x i64> %q0, i64 %q, i32 1
ret <2 x i64> %q1
}
-; CHECK: vpbroadcastq (%
+; CHECK: vbroadcastsd (%
define <4 x i64> @QQ64(i64* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load i64* %ptr, align 4
+ %q = load i64, i64* %ptr, align 4
%q0 = insertelement <4 x i64> undef, i64 %q, i32 0
%q1 = insertelement <4 x i64> %q0, i64 %q, i32 1
%q2 = insertelement <4 x i64> %q1, i64 %q, i32 2
; this used to crash
define <2 x double> @I(double* %ptr) nounwind uwtable readnone ssp {
entry:
- %q = load double* %ptr, align 4
+ %q = load double, double* %ptr, align 4
%vecinit.i = insertelement <2 x double> undef, double %q, i32 0
%vecinit2.i = insertelement <2 x double> %vecinit.i, double %q, i32 1
ret <2 x double> %vecinit2.i
;CHECK-LABEL: _inreg4xi64:
-;CHECK: vpbroadcastq
+;CHECK: vbroadcastsd
;CHECK: ret
define <4 x i64> @_inreg4xi64(<4 x i64> %a) {
%b = shufflevector <4 x i64> %a, <4 x i64> undef, <4 x i32> zeroinitializer
}
;CHECK-LABEL: _inreg2xdouble:
-;CHECK: vpbroadcastq
+;CHECK: vmovddup
;CHECK: ret
define <2 x double> @_inreg2xdouble(<2 x double> %a) {
%b = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> zeroinitializer
}
;CHECK-LABEL: _inreg8xi32:
-;CHECK: vpbroadcastd
+;CHECK: vbroadcastss
;CHECK: ret
define <8 x i32> @_inreg8xi32(<8 x i32> %a) {
%b = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> zeroinitializer
}
;CHECK-LABEL: _inreg4xi32:
-;CHECK: vpbroadcastd
+;CHECK: vbroadcastss
;CHECK: ret
define <4 x i32> @_inreg4xi32(<4 x i32> %a) {
%b = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> zeroinitializer
%b = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> zeroinitializer
ret <16 x i8> %b
}
+
+; These tests check that a vbroadcast instruction is used when we have a splat
+; formed from a concat_vectors (via the shufflevector) of two BUILD_VECTORs
+; (via the insertelements).
+
+; CHECK-LABEL: splat_concat1
+; CHECK-NOT: vinsertf128
+; CHECK: vbroadcastss
+; CHECK-NEXT: ret
+define <8 x float> @splat_concat1(float %f) {
+ %1 = insertelement <4 x float> undef, float %f, i32 0
+ %2 = insertelement <4 x float> %1, float %f, i32 1
+ %3 = insertelement <4 x float> %2, float %f, i32 2
+ %4 = insertelement <4 x float> %3, float %f, i32 3
+ %5 = shufflevector <4 x float> %4, <4 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ ret <8 x float> %5
+}
+
+; CHECK-LABEL: splat_concat2
+; CHECK-NOT: vinsertf128
+; CHECK: vbroadcastss
+; CHECK-NEXT: ret
+define <8 x float> @splat_concat2(float %f) {
+ %1 = insertelement <4 x float> undef, float %f, i32 0
+ %2 = insertelement <4 x float> %1, float %f, i32 1
+ %3 = insertelement <4 x float> %2, float %f, i32 2
+ %4 = insertelement <4 x float> %3, float %f, i32 3
+ %5 = insertelement <4 x float> undef, float %f, i32 0
+ %6 = insertelement <4 x float> %5, float %f, i32 1
+ %7 = insertelement <4 x float> %6, float %f, i32 2
+ %8 = insertelement <4 x float> %7, float %f, i32 3
+ %9 = shufflevector <4 x float> %4, <4 x float> %8, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ ret <8 x float> %9
+}
+
+; CHECK-LABEL: splat_concat3
+; CHECK-NOT: vinsertf128
+; CHECK: vbroadcastsd
+; CHECK-NEXT: ret
+define <4 x double> @splat_concat3(double %d) {
+ %1 = insertelement <2 x double> undef, double %d, i32 0
+ %2 = insertelement <2 x double> %1, double %d, i32 1
+ %3 = shufflevector <2 x double> %2, <2 x double> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ ret <4 x double> %3
+}
+
+; CHECK-LABEL: splat_concat4
+; CHECK-NOT: vinsertf128
+; CHECK: vbroadcastsd
+; CHECK-NEXT: ret
+define <4 x double> @splat_concat4(double %d) {
+ %1 = insertelement <2 x double> undef, double %d, i32 0
+ %2 = insertelement <2 x double> %1, double %d, i32 1
+ %3 = insertelement <2 x double> undef, double %d, i32 0
+ %4 = insertelement <2 x double> %3, double %d, i32 1
+ %5 = shufflevector <2 x double> %2, <2 x double> %4, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ ret <4 x double> %5
+}
+
+; Test cases for <rdar://problem/16074331>.
+; Instruction selection for broacast instruction fails if
+; the load cannot be folded into the broadcast.
+; This happens if the load has initial one use but other uses are
+; created later, or if selection DAG cannot prove that folding the
+; load will not create a cycle in the DAG.
+; Those test cases exerce the latter.
+
+; CHECK-LABEL: isel_crash_16b
+; CHECK: vpbroadcastb {{[^,]+}}, %xmm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_16b(i8* %cV_R.addr) {
+eintry:
+ %__a.addr.i = alloca <2 x i64>, align 16
+ %__b.addr.i = alloca <2 x i64>, align 16
+ %vCr = alloca <2 x i64>, align 16
+ store <2 x i64> zeroinitializer, <2 x i64>* %vCr, align 16
+ %tmp = load <2 x i64>, <2 x i64>* %vCr, align 16
+ %tmp2 = load i8, i8* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <16 x i8> undef, i8 %tmp2, i32 0
+ %splat.splat = shufflevector <16 x i8> %splat.splatinsert, <16 x i8> undef, <16 x i32> zeroinitializer
+ %tmp3 = bitcast <16 x i8> %splat.splat to <2 x i64>
+ store <2 x i64> %tmp, <2 x i64>* %__a.addr.i, align 16
+ store <2 x i64> %tmp3, <2 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_32b
+; CHECK: vpbroadcastb {{[^,]+}}, %ymm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_32b(i8* %cV_R.addr) {
+eintry:
+ %__a.addr.i = alloca <4 x i64>, align 16
+ %__b.addr.i = alloca <4 x i64>, align 16
+ %vCr = alloca <4 x i64>, align 16
+ store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
+ %tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
+ %tmp2 = load i8, i8* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <32 x i8> undef, i8 %tmp2, i32 0
+ %splat.splat = shufflevector <32 x i8> %splat.splatinsert, <32 x i8> undef, <32 x i32> zeroinitializer
+ %tmp3 = bitcast <32 x i8> %splat.splat to <4 x i64>
+ store <4 x i64> %tmp, <4 x i64>* %__a.addr.i, align 16
+ store <4 x i64> %tmp3, <4 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_8w
+; CHECK: vpbroadcastw {{[^,]+}}, %xmm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_8w(i16* %cV_R.addr) {
+entry:
+ %__a.addr.i = alloca <2 x i64>, align 16
+ %__b.addr.i = alloca <2 x i64>, align 16
+ %vCr = alloca <2 x i64>, align 16
+ store <2 x i64> zeroinitializer, <2 x i64>* %vCr, align 16
+ %tmp = load <2 x i64>, <2 x i64>* %vCr, align 16
+ %tmp2 = load i16, i16* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <8 x i16> undef, i16 %tmp2, i32 0
+ %splat.splat = shufflevector <8 x i16> %splat.splatinsert, <8 x i16> undef, <8 x i32> zeroinitializer
+ %tmp3 = bitcast <8 x i16> %splat.splat to <2 x i64>
+ store <2 x i64> %tmp, <2 x i64>* %__a.addr.i, align 16
+ store <2 x i64> %tmp3, <2 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_16w
+; CHECK: vpbroadcastw {{[^,]+}}, %ymm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_16w(i16* %cV_R.addr) {
+eintry:
+ %__a.addr.i = alloca <4 x i64>, align 16
+ %__b.addr.i = alloca <4 x i64>, align 16
+ %vCr = alloca <4 x i64>, align 16
+ store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
+ %tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
+ %tmp2 = load i16, i16* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <16 x i16> undef, i16 %tmp2, i32 0
+ %splat.splat = shufflevector <16 x i16> %splat.splatinsert, <16 x i16> undef, <16 x i32> zeroinitializer
+ %tmp3 = bitcast <16 x i16> %splat.splat to <4 x i64>
+ store <4 x i64> %tmp, <4 x i64>* %__a.addr.i, align 16
+ store <4 x i64> %tmp3, <4 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_4d
+; CHECK: vbroadcastss {{[^,]+}}, %xmm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_4d(i32* %cV_R.addr) {
+entry:
+ %__a.addr.i = alloca <2 x i64>, align 16
+ %__b.addr.i = alloca <2 x i64>, align 16
+ %vCr = alloca <2 x i64>, align 16
+ store <2 x i64> zeroinitializer, <2 x i64>* %vCr, align 16
+ %tmp = load <2 x i64>, <2 x i64>* %vCr, align 16
+ %tmp2 = load i32, i32* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <4 x i32> undef, i32 %tmp2, i32 0
+ %splat.splat = shufflevector <4 x i32> %splat.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
+ %tmp3 = bitcast <4 x i32> %splat.splat to <2 x i64>
+ store <2 x i64> %tmp, <2 x i64>* %__a.addr.i, align 16
+ store <2 x i64> %tmp3, <2 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_8d
+; CHECK: vbroadcastss {{[^,]+}}, %ymm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_8d(i32* %cV_R.addr) {
+eintry:
+ %__a.addr.i = alloca <4 x i64>, align 16
+ %__b.addr.i = alloca <4 x i64>, align 16
+ %vCr = alloca <4 x i64>, align 16
+ store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
+ %tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
+ %tmp2 = load i32, i32* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <8 x i32> undef, i32 %tmp2, i32 0
+ %splat.splat = shufflevector <8 x i32> %splat.splatinsert, <8 x i32> undef, <8 x i32> zeroinitializer
+ %tmp3 = bitcast <8 x i32> %splat.splat to <4 x i64>
+ store <4 x i64> %tmp, <4 x i64>* %__a.addr.i, align 16
+ store <4 x i64> %tmp3, <4 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_2q
+; CHECK: vpbroadcastq {{[^,]+}}, %xmm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_2q(i64* %cV_R.addr) {
+entry:
+ %__a.addr.i = alloca <2 x i64>, align 16
+ %__b.addr.i = alloca <2 x i64>, align 16
+ %vCr = alloca <2 x i64>, align 16
+ store <2 x i64> zeroinitializer, <2 x i64>* %vCr, align 16
+ %tmp = load <2 x i64>, <2 x i64>* %vCr, align 16
+ %tmp2 = load i64, i64* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <2 x i64> undef, i64 %tmp2, i32 0
+ %splat.splat = shufflevector <2 x i64> %splat.splatinsert, <2 x i64> undef, <2 x i32> zeroinitializer
+ store <2 x i64> %tmp, <2 x i64>* %__a.addr.i, align 16
+ store <2 x i64> %splat.splat, <2 x i64>* %__b.addr.i, align 16
+ ret void
+}
+
+; CHECK-LABEL: isel_crash_4q
+; CHECK: vbroadcastsd {{[^,]+}}, %ymm{{[0-9]+}}
+; CHECK: ret
+define void @isel_crash_4q(i64* %cV_R.addr) {
+eintry:
+ %__a.addr.i = alloca <4 x i64>, align 16
+ %__b.addr.i = alloca <4 x i64>, align 16
+ %vCr = alloca <4 x i64>, align 16
+ store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
+ %tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
+ %tmp2 = load i64, i64* %cV_R.addr, align 4
+ %splat.splatinsert = insertelement <4 x i64> undef, i64 %tmp2, i32 0
+ %splat.splat = shufflevector <4 x i64> %splat.splatinsert, <4 x i64> undef, <4 x i32> zeroinitializer
+ store <4 x i64> %tmp, <4 x i64>* %__a.addr.i, align 16
+ store <4 x i64> %splat.splat, <4 x i64>* %__b.addr.i, align 16
+ ret void
+}