PTX: add test cases for cvt, fneg, and selp

Patch by Dan Bailey

llvm-svn: 131128

3 files changed, 274 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/PTX/cvt.ll b/llvm/test/CodeGen/PTX/cvt.ll
new file mode 100644
index 00000000000..2f793dede62
--- /dev/null
+++ b/llvm/test/CodeGen/PTX/cvt.ll
@@ -0,0 +1,234 @@
+; RUN: llc < %s -march=ptx32 | FileCheck %s
+
+; preds 
+; (note: we convert back to i32 to return)
+
+define ptx_device i32 @cvt_pred_i16(i16 %x, i1 %y) {
+; CHECK: cvt.pred.u16 p0, rh1;
+; CHECK: ret;
+	%a = trunc i16 %x to i1
+	%b = and i1 %a, %y
+	%c = zext i1 %b to i32
+	ret i32 %c
+}
+
+define ptx_device i32 @cvt_pred_i32(i32 %x, i1 %y) {
+; CHECK: cvt.pred.u32 p0, r1;
+; CHECK: ret;
+	%a = trunc i32 %x to i1
+	%b = and i1 %a, %y
+	%c = zext i1 %b to i32
+	ret i32 %c
+}
+
+define ptx_device i32 @cvt_pred_i64(i64 %x, i1 %y) {
+; CHECK: cvt.pred.u64 p0, rd1;
+; CHECK: ret;
+	%a = trunc i64 %x to i1
+	%b = and i1 %a, %y
+	%c = zext i1 %b to i32
+	ret i32 %c
+}
+
+define ptx_device i32 @cvt_pred_f32(float %x, i1 %y) {
+; CHECK: cvt.rni.pred.f32 p0, f1;
+; CHECK: ret;
+	%a = fptoui float %x to i1
+	%b = and i1 %a, %y
+	%c = zext i1 %b to i32
+	ret i32 %c
+}
+
+define ptx_device i32 @cvt_pred_f64(double %x, i1 %y) {
+; CHECK: cvt.rni.pred.f64 p0, fd1;
+; CHECK: ret;
+	%a = fptoui double %x to i1
+	%b = and i1 %a, %y
+	%c = zext i1 %b to i32
+	ret i32 %c
+}
+
+; i16
+
+define ptx_device i16 @cvt_i16_preds(i1 %x) {
+; CHECK: cvt.u16.pred rh0, p1;
+; CHECK: ret;
+	%a = zext i1 %x to i16
+	ret i16 %a
+}
+
+define ptx_device i16 @cvt_i16_i32(i32 %x) {
+; CHECK: cvt.u16.u32 rh0, r1;
+; CHECK: ret;
+	%a = trunc i32 %x to i16
+	ret i16 %a
+}
+
+define ptx_device i16 @cvt_i16_i64(i64 %x) {
+; CHECK: cvt.u16.u64 rh0, rd1;
+; CHECK: ret;
+	%a = trunc i64 %x to i16
+	ret i16 %a
+}
+
+define ptx_device i16 @cvt_i16_f32(float %x) {
+; CHECK: cvt.rni.u16.f32 rh0, f1;
+; CHECK: ret;
+	%a = fptoui float %x to i16
+	ret i16 %a
+}
+
+define ptx_device i16 @cvt_i16_f64(double %x) {
+; CHECK: cvt.rni.u16.f64 rh0, fd1;
+; CHECK: ret;
+	%a = fptoui double %x to i16
+	ret i16 %a
+}
+
+; i32
+
+define ptx_device i32 @cvt_i32_preds(i1 %x) {
+; CHECK: cvt.u32.pred r0, p1;
+; CHECK: ret;
+	%a = zext i1 %x to i32
+	ret i32 %a
+}
+
+define ptx_device i32 @cvt_i32_i16(i16 %x) {
+; CHECK: cvt.u32.u16 r0, rh1;
+; CHECK: ret;
+	%a = zext i16 %x to i32
+	ret i32 %a
+}
+
+define ptx_device i32 @cvt_i32_i64(i64 %x) {
+; CHECK: cvt.u32.u64 r0, rd1;
+; CHECK: ret;
+	%a = trunc i64 %x to i32
+	ret i32 %a
+}
+
+define ptx_device i32 @cvt_i32_f32(float %x) {
+; CHECK: cvt.rni.u32.f32 r0, f1;
+; CHECK: ret;
+	%a = fptoui float %x to i32
+	ret i32 %a
+}
+
+define ptx_device i32 @cvt_i32_f64(double %x) {
+; CHECK: cvt.rni.u32.f64 r0, fd1;
+; CHECK: ret;
+	%a = fptoui double %x to i32
+	ret i32 %a
+}
+
+; i64
+
+define ptx_device i64 @cvt_i64_preds(i1 %x) {
+; CHECK: cvt.u64.pred rd0, p1;
+; CHECK: ret;
+	%a = zext i1 %x to i64
+	ret i64 %a
+}
+
+define ptx_device i64 @cvt_i64_i16(i16 %x) {
+; CHECK: cvt.u64.u16 rd0, rh1;
+; CHECK: ret;
+	%a = zext i16 %x to i64
+	ret i64 %a
+}
+
+define ptx_device i64 @cvt_i64_i32(i32 %x) {
+; CHECK: cvt.u64.u32 rd0, r1;
+; CHECK: ret;
+	%a = zext i32 %x to i64
+	ret i64 %a
+}
+
+define ptx_device i64 @cvt_i64_f32(float %x) {
+; CHECK: cvt.rni.u64.f32 rd0, f1;
+; CHECK: ret;
+	%a = fptoui float %x to i64
+	ret i64 %a
+}
+
+define ptx_device i64 @cvt_i64_f64(double %x) {
+; CHECK: cvt.rni.u64.f64 rd0, fd1;
+; CHECK: ret;
+	%a = fptoui double %x to i64
+	ret i64 %a
+}
+
+; f32
+
+define ptx_device float @cvt_f32_preds(i1 %x) {
+; CHECK: cvt.rn.f32.pred f0, p1;
+; CHECK: ret;
+	%a = uitofp i1 %x to float
+	ret float %a
+}
+
+define ptx_device float @cvt_f32_i16(i16 %x) {
+; CHECK: cvt.rn.f32.u16 f0, rh1;
+; CHECK: ret;
+	%a = uitofp i16 %x to float
+	ret float %a
+}
+
+define ptx_device float @cvt_f32_i32(i32 %x) {
+; CHECK: cvt.rn.f32.u32 f0, r1;
+; CHECK: ret;
+	%a = uitofp i32 %x to float
+	ret float %a
+}
+
+define ptx_device float @cvt_f32_i64(i64 %x) {
+; CHECK: cvt.rn.f32.u64 f0, rd1;
+; CHECK: ret;
+	%a = uitofp i64 %x to float
+	ret float %a
+}
+
+define ptx_device float @cvt_f32_f64(double %x) {
+; CHECK: cvt.rn.f32.f64 f0, fd1;
+; CHECK: ret;
+	%a = fptrunc double %x to float
+	ret float %a
+}
+
+; f64
+
+define ptx_device double @cvt_f64_preds(i1 %x) {
+; CHECK: cvt.rn.f64.pred fd0, p1;
+; CHECK: ret;
+	%a = uitofp i1 %x to double
+	ret double %a
+}
+
+define ptx_device double @cvt_f64_i16(i16 %x) {
+; CHECK: cvt.rn.f64.u16 fd0, rh1;
+; CHECK: ret;
+	%a = uitofp i16 %x to double
+	ret double %a
+}
+
+define ptx_device double @cvt_f64_i32(i32 %x) {
+; CHECK: cvt.rn.f64.u32 fd0, r1;
+; CHECK: ret;
+	%a = uitofp i32 %x to double
+	ret double %a
+}
+
+define ptx_device double @cvt_f64_i64(i64 %x) {
+; CHECK: cvt.rn.f64.u64 fd0, rd1;
+; CHECK: ret;
+	%a = uitofp i64 %x to double
+	ret double %a
+}
+
+define ptx_device double @cvt_f64_f32(float %x) {
+; CHECK: cvt.f64.f32 fd0, f1;
+; CHECK: ret;
+	%a = fpext float %x to double
+	ret double %a
+}
diff --git a/llvm/test/CodeGen/PTX/fneg.ll b/llvm/test/CodeGen/PTX/fneg.ll
new file mode 100644
index 00000000000..22eeda3f0cb
--- /dev/null
+++ b/llvm/test/CodeGen/PTX/fneg.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -march=ptx32 | FileCheck %s
+
+define ptx_device float @t1_f32(float %x) {
+; CHECK: neg.f32 f0, f1;
+; CHECK-NEXT: ret;
+	%y = fsub float -0.000000e+00, %x
+	ret float %y
+}
+
+define ptx_device double @t1_f64(double %x) {
+; CHECK: neg.f64 fd0, fd1;
+; CHECK-NEXT: ret;
+	%y = fsub double -0.000000e+00, %x
+	ret double %y
+}
diff --git a/llvm/test/CodeGen/PTX/selp.ll b/llvm/test/CodeGen/PTX/selp.ll
new file mode 100644
index 00000000000..6f1b03e5997
--- /dev/null
+++ b/llvm/test/CodeGen/PTX/selp.ll
@@ -0,0 +1,25 @@
+; RUN: llc < %s -march=ptx32 | FileCheck %s
+
+define ptx_device i32 @test_selp_i32(i1 %x, i32 %y, i32 %z) {
+; CHECK: selp.u32 r0, r1, r2, p1;
+	%a = select i1 %x, i32 %y, i32 %z
+	ret i32 %a
+}
+
+define ptx_device i64 @test_selp_i64(i1 %x, i64 %y, i64 %z) {
+; CHECK: selp.u64 rd0, rd1, rd2, p1;
+	%a = select i1 %x, i64 %y, i64 %z
+	ret i64 %a
+}
+
+define ptx_device float @test_selp_f32(i1 %x, float %y, float %z) {
+; CHECK: selp.f32 f0, f1, f2, p1;
+	%a = select i1 %x, float %y, float %z
+	ret float %a
+}
+
+define ptx_device double @test_selp_f64(i1 %x, double %y, double %z) {
+; CHECK: selp.f64 fd0, fd1, fd2, p1;
+	%a = select i1 %x, double %y, double %z
+	ret double %a
+}