Extend initial support for primitive types in PTX backend

- Allow i16, i32, i64, float, and double types, using the native .u16,
  .u32, .u64, .f32, and .f64 PTX types.
- Allow loading/storing of all primitive types.
- Allow primitive types to be passed as parameters.
- Allow selection of PTX Version and Shader Model as sub-target attributes.
- Merge integer/floating-point test cases for load/store.
- Use .u32 instead of .s32 to conform to output from NVidia nvcc compiler.

Patch by Justin Holewinski



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@126824 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 883 insertions, 254 deletions

diff --git a/test/CodeGen/PTX/add.ll b/test/CodeGen/PTX/add.ll
index 9e777ae30c..598591c0fc 100644
--- a/test/CodeGen/PTX/add.ll
+++ b/test/CodeGen/PTX/add.ll
@@ -1,29 +1,71 @@
 ; RUN: llc < %s -march=ptx | FileCheck %s
 
-define ptx_device i32 @t1(i32 %x, i32 %y) {
-; CHECK: add.s32 r0, r1, r2;
+define ptx_device i16 @t1_u16(i16 %x, i16 %y) {
+; CHECK: add.u16 rh0, rh1, rh2;
+; CHECK-NEXT: ret;
+	%z = add i16 %x, %y
+	ret i16 %z
+}
+
+define ptx_device i32 @t1_u32(i32 %x, i32 %y) {
+; CHECK: add.u32 r0, r1, r2;
+; CHECK-NEXT: ret;
 	%z = add i32 %x, %y
-; CHECK: ret;
 	ret i32 %z
 }
 
-define ptx_device i32 @t2(i32 %x) {
-; CHECK: add.s32 r0, r1, 1;
-	%z = add i32 %x, 1
-; CHECK: ret;
-	ret i32 %z
+define ptx_device i64 @t1_u64(i64 %x, i64 %y) {
+; CHECK: add.u64 rd0, rd1, rd2;
+; CHECK-NEXT: ret;
+	%z = add i64 %x, %y
+	ret i64 %z
 }
 
-define ptx_device float @t3(float %x, float %y) {
+define ptx_device float @t1_f32(float %x, float %y) {
 ; CHECK: add.f32 f0, f1, f2
 ; CHECK-NEXT: ret;
   %z = fadd float %x, %y
   ret float %z
 }
 
-define ptx_device float @t4(float %x) {
+define ptx_device double @t1_f64(double %x, double %y) {
+; CHECK: add.f64 fd0, fd1, fd2
+; CHECK-NEXT: ret;
+  %z = fadd double %x, %y
+  ret double %z
+}
+
+define ptx_device i16 @t2_u16(i16 %x) {
+; CHECK: add.u16 rh0, rh1, 1;
+; CHECK-NEXT: ret;
+	%z = add i16 %x, 1
+	ret i16 %z
+}
+
+define ptx_device i32 @t2_u32(i32 %x) {
+; CHECK: add.u32 r0, r1, 1;
+; CHECK-NEXT: ret;
+	%z = add i32 %x, 1
+	ret i32 %z
+}
+
+define ptx_device i64 @t2_u64(i64 %x) {
+; CHECK: add.u64 rd0, rd1, 1;
+; CHECK-NEXT: ret;
+	%z = add i64 %x, 1
+	ret i64 %z
+}
+
+define ptx_device float @t2_f32(float %x) {
 ; CHECK: add.f32 f0, f1, 0F3F800000;
 ; CHECK-NEXT: ret;
   %z = fadd float %x, 1.0
   ret float %z
 }
+
+define ptx_device double @t2_f64(double %x) {
+; CHECK: add.f64 fd0, fd1, 0D3FF0000000000000;
+; CHECK-NEXT: ret;
+  %z = fadd double %x, 1.0
+  ret double %z
+}
diff --git a/test/CodeGen/PTX/ld.ll b/test/CodeGen/PTX/ld.ll
index 836c4d4104..e7cc92e3c9 100644
--- a/test/CodeGen/PTX/ld.ll
+++ b/test/CodeGen/PTX/ld.ll
@@ -1,78 +1,422 @@
 ; RUN: llc < %s -march=ptx | FileCheck %s
 
-;CHECK: .extern .global .s32 array[];
-@array = external global [10 x i32]
+;CHECK: .extern .global .u16 array_i16[];
+@array_i16 = external global [10 x i16]
 
-;CHECK: .extern .const .s32 array_constant[];
-@array_constant = external addrspace(1) constant [10 x i32]
+;CHECK: .extern .const .u16 array_constant_i16[];
+@array_constant_i16 = external addrspace(1) constant [10 x i16]
 
-;CHECK: .extern .local .s32 array_local[];
-@array_local = external addrspace(2) global [10 x i32]
+;CHECK: .extern .local .u16 array_local_i16[];
+@array_local_i16 = external addrspace(2) global [10 x i16]
 
-;CHECK: .extern .shared .s32 array_shared[];
-@array_shared = external addrspace(4) global [10 x i32]
+;CHECK: .extern .shared .u16 array_shared_i16[];
+@array_shared_i16 = external addrspace(4) global [10 x i16]
 
-define ptx_device i32 @t1(i32* %p) {
+;CHECK: .extern .global .u32 array_i32[];
+@array_i32 = external global [10 x i32]
+
+;CHECK: .extern .const .u32 array_constant_i32[];
+@array_constant_i32 = external addrspace(1) constant [10 x i32]
+
+;CHECK: .extern .local .u32 array_local_i32[];
+@array_local_i32 = external addrspace(2) global [10 x i32]
+
+;CHECK: .extern .shared .u32 array_shared_i32[];
+@array_shared_i32 = external addrspace(4) global [10 x i32]
+
+;CHECK: .extern .global .u64 array_i64[];
+@array_i64 = external global [10 x i64]
+
+;CHECK: .extern .const .u64 array_constant_i64[];
+@array_constant_i64 = external addrspace(1) constant [10 x i64]
+
+;CHECK: .extern .local .u64 array_local_i64[];
+@array_local_i64 = external addrspace(2) global [10 x i64]
+
+;CHECK: .extern .shared .u64 array_shared_i64[];
+@array_shared_i64 = external addrspace(4) global [10 x i64]
+
+;CHECK: .extern .global .f32 array_float[];
+@array_float = external global [10 x float]
+
+;CHECK: .extern .const .f32 array_constant_float[];
+@array_constant_float = external addrspace(1) constant [10 x float]
+
+;CHECK: .extern .local .f32 array_local_float[];
+@array_local_float = external addrspace(2) global [10 x float]
+
+;CHECK: .extern .shared .f32 array_shared_float[];
+@array_shared_float = external addrspace(4) global [10 x float]
+
+;CHECK: .extern .global .f64 array_double[];
+@array_double = external global [10 x double]
+
+;CHECK: .extern .const .f64 array_constant_double[];
+@array_constant_double = external addrspace(1) constant [10 x double]
+
+;CHECK: .extern .local .f64 array_local_double[];
+@array_local_double = external addrspace(2) global [10 x double]
+
+;CHECK: .extern .shared .f64 array_shared_double[];
+@array_shared_double = external addrspace(4) global [10 x double]
+
+
+define ptx_device i16 @t1_u16(i16* %p) {
 entry:
-;CHECK: ld.global.s32 r0, [r1];
+;CHECK: ld.global.u16 rh0, [r1];
+;CHECK-NEXT; ret;
+  %x = load i16* %p
+  ret i16 %x
+}
+
+define ptx_device i32 @t1_u32(i32* %p) {
+entry:
+;CHECK: ld.global.u32 r0, [r1];
+;CHECK-NEXT: ret;
   %x = load i32* %p
   ret i32 %x
 }
 
-define ptx_device i32 @t2(i32* %p) {
+define ptx_device i64 @t1_u64(i64* %p) {
+entry:
+;CHECK: ld.global.u64 rd0, [r1];
+;CHECK-NEXT: ret;
+  %x = load i64* %p
+  ret i64 %x
+}
+
+define ptx_device float @t1_f32(float* %p) {
+entry:
+;CHECK: ld.global.f32 f0, [r1];
+;CHECK-NEXT: ret;
+  %x = load float* %p
+  ret float %x
+}
+
+define ptx_device double @t1_f64(double* %p) {
+entry:
+;CHECK: ld.global.f64 fd0, [r1];
+;CHECK-NEXT: ret;
+  %x = load double* %p
+  ret double %x
+}
+
+define ptx_device i16 @t2_u16(i16* %p) {
 entry:
-;CHECK: ld.global.s32 r0, [r1+4];
+;CHECK: ld.global.u16 rh0, [r1+2];
+;CHECK-NEXT: ret;
+  %i = getelementptr i16* %p, i32 1
+  %x = load i16* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t2_u32(i32* %p) {
+entry:
+;CHECK: ld.global.u32 r0, [r1+4];
+;CHECK-NEXT: ret;
   %i = getelementptr i32* %p, i32 1
   %x = load i32* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t3(i32* %p, i32 %q) {
+define ptx_device i64 @t2_u64(i64* %p) {
+entry:
+;CHECK: ld.global.u64 rd0, [r1+8];
+;CHECK-NEXT: ret;
+  %i = getelementptr i64* %p, i32 1
+  %x = load i64* %i
+  ret i64 %x
+}
+
+define ptx_device float @t2_f32(float* %p) {
+entry:
+;CHECK: ld.global.f32 f0, [r1+4];
+;CHECK-NEXT: ret;
+  %i = getelementptr float* %p, i32 1
+  %x = load float* %i
+  ret float %x
+}
+
+define ptx_device double @t2_f64(double* %p) {
+entry:
+;CHECK: ld.global.f64 fd0, [r1+8];
+;CHECK-NEXT: ret;
+  %i = getelementptr double* %p, i32 1
+  %x = load double* %i
+  ret double %x
+}
+
+define ptx_device i16 @t3_u16(i16* %p, i32 %q) {
+entry:
+;CHECK: shl.b32 r0, r2, 1;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: ld.global.u16 rh0, [r0];
+  %i = getelementptr i16* %p, i32 %q
+  %x = load i16* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t3_u32(i32* %p, i32 %q) {
 entry:
 ;CHECK: shl.b32 r0, r2, 2;
-;CHECK: add.s32 r0, r1, r0;
-;CHECK: ld.global.s32 r0, [r0];
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: ld.global.u32 r0, [r0];
   %i = getelementptr i32* %p, i32 %q
   %x = load i32* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t4_global() {
+define ptx_device i64 @t3_u64(i64* %p, i32 %q) {
+entry:
+;CHECK: shl.b32 r0, r2, 3;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: ld.global.u64 rd0, [r0];
+  %i = getelementptr i64* %p, i32 %q
+  %x = load i64* %i
+  ret i64 %x
+}
+
+define ptx_device float @t3_f32(float* %p, i32 %q) {
+entry:
+;CHECK: shl.b32 r0, r2, 2;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: ld.global.f32 f0, [r0];
+  %i = getelementptr float* %p, i32 %q
+  %x = load float* %i
+  ret float %x
+}
+
+define ptx_device double @t3_f64(double* %p, i32 %q) {
+entry:
+;CHECK: shl.b32 r0, r2, 3;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: ld.global.f64 fd0, [r0];
+  %i = getelementptr double* %p, i32 %q
+  %x = load double* %i
+  ret double %x
+}
+
+define ptx_device i16 @t4_global_u16() {
+entry:
+;CHECK: ld.global.u16 rh0, [array_i16];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16]* @array_i16, i32 0, i32 0
+  %x = load i16* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t4_global_u32() {
 entry:
-;CHECK: ld.global.s32 r0, [array];
-  %i = getelementptr [10 x i32]* @array, i32 0, i32 0
+;CHECK: ld.global.u32 r0, [array_i32];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32]* @array_i32, i32 0, i32 0
   %x = load i32* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t4_const() {
+define ptx_device i64 @t4_global_u64() {
 entry:
-;CHECK: ld.const.s32 r0, [array_constant];
-  %i = getelementptr [10 x i32] addrspace(1)* @array_constant, i32 0, i32 0
+;CHECK: ld.global.u64 rd0, [array_i64];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64]* @array_i64, i32 0, i32 0
+  %x = load i64* %i
+  ret i64 %x
+}
+
+define ptx_device float @t4_global_f32() {
+entry:
+;CHECK: ld.global.f32 f0, [array_float];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float]* @array_float, i32 0, i32 0
+  %x = load float* %i
+  ret float %x
+}
+
+define ptx_device double @t4_global_f64() {
+entry:
+;CHECK: ld.global.f64 fd0, [array_double];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double]* @array_double, i32 0, i32 0
+  %x = load double* %i
+  ret double %x
+}
+
+define ptx_device i16 @t4_const_u16() {
+entry:
+;CHECK: ld.const.u16 rh0, [array_constant_i16];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16] addrspace(1)* @array_constant_i16, i32 0, i32 0
+  %x = load i16 addrspace(1)* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t4_const_u32() {
+entry:
+;CHECK: ld.const.u32 r0, [array_constant_i32];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32] addrspace(1)* @array_constant_i32, i32 0, i32 0
   %x = load i32 addrspace(1)* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t4_local() {
+define ptx_device i64 @t4_const_u64() {
+entry:
+;CHECK: ld.const.u64 rd0, [array_constant_i64];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64] addrspace(1)* @array_constant_i64, i32 0, i32 0
+  %x = load i64 addrspace(1)* %i
+  ret i64 %x
+}
+
+define ptx_device float @t4_const_f32() {
 entry:
-;CHECK: ld.local.s32 r0, [array_local];
-  %i = getelementptr [10 x i32] addrspace(2)* @array_local, i32 0, i32 0
+;CHECK: ld.const.f32 f0, [array_constant_float];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float] addrspace(1)* @array_constant_float, i32 0, i32 0
+  %x = load float addrspace(1)* %i
+  ret float %x
+}
+
+define ptx_device double @t4_const_f64() {
+entry:
+;CHECK: ld.const.f64 fd0, [array_constant_double];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double] addrspace(1)* @array_constant_double, i32 0, i32 0
+  %x = load double addrspace(1)* %i
+  ret double %x
+}
+
+define ptx_device i16 @t4_local_u16() {
+entry:
+;CHECK: ld.local.u16 rh0, [array_local_i16];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16] addrspace(2)* @array_local_i16, i32 0, i32 0
+  %x = load i16 addrspace(2)* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t4_local_u32() {
+entry:
+;CHECK: ld.local.u32 r0, [array_local_i32];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32] addrspace(2)* @array_local_i32, i32 0, i32 0
   %x = load i32 addrspace(2)* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t4_shared() {
+define ptx_device i64 @t4_local_u64() {
 entry:
-;CHECK: ld.shared.s32 r0, [array_shared];
-  %i = getelementptr [10 x i32] addrspace(4)* @array_shared, i32 0, i32 0
+;CHECK: ld.local.u64 rd0, [array_local_i64];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64] addrspace(2)* @array_local_i64, i32 0, i32 0
+  %x = load i64 addrspace(2)* %i
+  ret i64 %x
+}
+
+define ptx_device float @t4_local_f32() {
+entry:
+;CHECK: ld.local.f32 f0, [array_local_float];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float] addrspace(2)* @array_local_float, i32 0, i32 0
+  %x = load float addrspace(2)* %i
+  ret float %x
+}
+
+define ptx_device double @t4_local_f64() {
+entry:
+;CHECK: ld.local.f64 fd0, [array_local_double];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double] addrspace(2)* @array_local_double, i32 0, i32 0
+  %x = load double addrspace(2)* %i
+  ret double %x
+}
+
+define ptx_device i16 @t4_shared_u16() {
+entry:
+;CHECK: ld.shared.u16 rh0, [array_shared_i16];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16] addrspace(4)* @array_shared_i16, i32 0, i32 0
+  %x = load i16 addrspace(4)* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t4_shared_u32() {
+entry:
+;CHECK: ld.shared.u32 r0, [array_shared_i32];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32] addrspace(4)* @array_shared_i32, i32 0, i32 0
   %x = load i32 addrspace(4)* %i
   ret i32 %x
 }
 
-define ptx_device i32 @t5() {
+define ptx_device i64 @t4_shared_u64() {
+entry:
+;CHECK: ld.shared.u64 rd0, [array_shared_i64];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64] addrspace(4)* @array_shared_i64, i32 0, i32 0
+  %x = load i64 addrspace(4)* %i
+  ret i64 %x
+}
+
+define ptx_device float @t4_shared_f32() {
+entry:
+;CHECK: ld.shared.f32 f0, [array_shared_float];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float] addrspace(4)* @array_shared_float, i32 0, i32 0
+  %x = load float addrspace(4)* %i
+  ret float %x
+}
+
+define ptx_device double @t4_shared_f64() {
+entry:
+;CHECK: ld.shared.f64 fd0, [array_shared_double];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double] addrspace(4)* @array_shared_double, i32 0, i32 0
+  %x = load double addrspace(4)* %i
+  ret double %x
+}
+
+define ptx_device i16 @t5_u16() {
 entry:
-;CHECK: ld.global.s32 r0, [array+4];
-  %i = getelementptr [10 x i32]* @array, i32 0, i32 1
+;CHECK: ld.global.u16 rh0, [array_i16+2];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16]* @array_i16, i32 0, i32 1
+  %x = load i16* %i
+  ret i16 %x
+}
+
+define ptx_device i32 @t5_u32() {
+entry:
+;CHECK: ld.global.u32 r0, [array_i32+4];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32]* @array_i32, i32 0, i32 1
   %x = load i32* %i
   ret i32 %x
 }
+
+define ptx_device i64 @t5_u64() {
+entry:
+;CHECK: ld.global.u64 rd0, [array_i64+8];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64]* @array_i64, i32 0, i32 1
+  %x = load i64* %i
+  ret i64 %x
+}
+
+define ptx_device float @t5_f32() {
+entry:
+;CHECK: ld.global.f32 f0, [array_float+4];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float]* @array_float, i32 0, i32 1
+  %x = load float* %i
+  ret float %x
+}
+
+define ptx_device double @t5_f64() {
+entry:
+;CHECK: ld.global.f64 fd0, [array_double+8];
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double]* @array_double, i32 0, i32 1
+  %x = load double* %i
+  ret double %x
+}
diff --git a/test/CodeGen/PTX/ld_float.ll b/test/CodeGen/PTX/ld_float.ll
deleted file mode 100644
index 62d2c36e64..0000000000
--- a/test/CodeGen/PTX/ld_float.ll
+++ /dev/null
@@ -1,86 +0,0 @@
-; RUN: llc < %s -march=ptx | FileCheck %s
-
-;CHECK: .extern .global .f32 array[];
-@array = external global [10 x float]
-
-;CHECK: .extern .const .f32 array_constant[];
-@array_constant = external addrspace(1) constant [10 x float]
-
-;CHECK: .extern .local .f32 array_local[];
-@array_local = external addrspace(2) global [10 x float]
-
-;CHECK: .extern .shared .f32 array_shared[];
-@array_shared = external addrspace(4) global [10 x float]
-
-define ptx_device float @t1(float* %p) {
-entry:
-;CHECK: ld.global.f32 f0, [r1];
-;CHECK-NEXT: ret;
-  %x = load float* %p
-  ret float %x
-}
-
-define ptx_device float @t2(float* %p) {
-entry:
-;CHECK: ld.global.f32 f0, [r1+4];
-;CHECK-NEXT: ret;
-  %i = getelementptr float* %p, i32 1
-  %x = load float* %i
-  ret float %x
-}
-
-define ptx_device float @t3(float* %p, i32 %q) {
-entry:
-;CHECK: shl.b32 r0, r2, 2;
-;CHECK-NEXT: add.s32 r0, r1, r0;
-;CHECK-NEXT: ld.global.f32 f0, [r0];
-;CHECK-NEXT: ret;
-  %i = getelementptr float* %p, i32 %q
-  %x = load float* %i
-  ret float %x
-}
-
-define ptx_device float @t4_global() {
-entry:
-;CHECK: ld.global.f32 f0, [array];
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float]* @array, i32 0, i32 0
-  %x = load float* %i
-  ret float %x
-}
-
-define ptx_device float @t4_const() {
-entry:
-;CHECK: ld.const.f32 f0, [array_constant];
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float] addrspace(1)* @array_constant, i32 0, i32 0
-  %x = load float addrspace(1)* %i
-  ret float %x
-}
-
-define ptx_device float @t4_local() {
-entry:
-;CHECK: ld.local.f32 f0, [array_local];
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float] addrspace(2)* @array_local, i32 0, i32 0
-  %x = load float addrspace(2)* %i
-  ret float %x
-}
-
-define ptx_device float @t4_shared() {
-entry:
-;CHECK: ld.shared.f32 f0, [array_shared];
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float] addrspace(4)* @array_shared, i32 0, i32 0
-  %x = load float addrspace(4)* %i
-  ret float %x
-}
-
-define ptx_device float @t5() {
-entry:
-;CHECK: ld.global.f32 f0, [array+4];
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float]* @array, i32 0, i32 1
-  %x = load float* %i
-  ret float %x
-}
diff --git a/test/CodeGen/PTX/mov.ll b/test/CodeGen/PTX/mov.ll
index d201a7867a..00dcf19f1d 100644
--- a/test/CodeGen/PTX/mov.ll
+++ b/test/CodeGen/PTX/mov.ll
@@ -1,25 +1,62 @@
 ; RUN: llc < %s -march=ptx | FileCheck %s
 
-define ptx_device i32 @t1() {
-; CHECK: mov.s32 r0, 0;
+define ptx_device i16 @t1_u16() {
+; CHECK: mov.u16 rh0, 0;
+; CHECK: ret;
+	ret i16 0
+}
+
+define ptx_device i32 @t1_u32() {
+; CHECK: mov.u32 r0, 0;
 ; CHECK: ret;
 	ret i32 0
 }
 
-define ptx_device i32 @t2(i32 %x) {
-; CHECK: mov.s32 r0, r1;
+define ptx_device i64 @t1_u64() {
+; CHECK: mov.u64 rd0, 0;
 ; CHECK: ret;
-	ret i32 %x
+	ret i64 0
 }
 
-define ptx_device float @t3() {
+define ptx_device float @t1_f32() {
 ; CHECK: mov.f32 f0, 0F00000000;
-; CHECK-NEXT: ret;
+; CHECK: ret;
 	ret float 0.0
 }
 
-define ptx_device float @t4(float %x) {
+define ptx_device double @t1_f64() {
+; CHECK: mov.f64 fd0, 0D0000000000000000;
+; CHECK: ret;
+	ret double 0.0
+}
+
+define ptx_device i16 @t2_u16(i16 %x) {
+; CHECK: mov.u16 rh0, rh1;
+; CHECK: ret;
+	ret i16 %x
+}
+
+define ptx_device i32 @t2_u32(i32 %x) {
+; CHECK: mov.u32 r0, r1;
+; CHECK: ret;
+	ret i32 %x
+}
+
+define ptx_device i64 @t2_u64(i64 %x) {
+; CHECK: mov.u64 rd0, rd1;
+; CHECK: ret;
+	ret i64 %x
+}
+
+define ptx_device float @t3_f32(float %x) {
 ; CHECK: mov.f32 f0, f1;
 ; CHECK-NEXT: ret;
 	ret float %x
 }
+
+define ptx_device double @t3_f64(double %x) {
+; CHECK: mov.f64 fd0, fd1;
+; CHECK-NEXT: ret;
+	ret double %x
+}
+
diff --git a/test/CodeGen/PTX/mul.ll b/test/CodeGen/PTX/mul.ll
index 01871da489..fd0788fce6 100644
--- a/test/CodeGen/PTX/mul.ll
+++ b/test/CodeGen/PTX/mul.ll
@@ -10,16 +10,30 @@
 ;	ret i32 %z
 ;}
 
-define ptx_device float @t3(float %x, float %y) {
+define ptx_device float @t1_f32(float %x, float %y) {
 ; CHECK: mul.f32 f0, f1, f2
 ; CHECK-NEXT: ret;
   %z = fmul float %x, %y
   ret float %z
 }
 
-define ptx_device float @t4(float %x) {
+define ptx_device double @t1_f64(double %x, double %y) {
+; CHECK: mul.f64 fd0, fd1, fd2
+; CHECK-NEXT: ret;
+  %z = fmul double %x, %y
+  ret double %z
+}
+
+define ptx_device float @t2_f32(float %x) {
 ; CHECK: mul.f32 f0, f1, 0F40A00000;
 ; CHECK-NEXT: ret;
   %z = fmul float %x, 5.0
   ret float %z
 }
+
+define ptx_device double @t2_f64(double %x) {
+; CHECK: mul.f64 fd0, fd1, 0D4014000000000000;
+; CHECK-NEXT: ret;
+  %z = fmul double %x, 5.0
+  ret double %z
+}
diff --git a/test/CodeGen/PTX/options.ll b/test/CodeGen/PTX/options.ll
index a14d5c9c27..1435537e00 100644
--- a/test/CodeGen/PTX/options.ll
+++ b/test/CodeGen/PTX/options.ll
@@ -1,5 +1,8 @@
-; RUN: llc < %s -march=ptx -ptx-version=2.0 | grep ".version 2.0"
-; RUN: llc < %s -march=ptx -ptx-target=sm_20 | grep ".target sm_20"
+; RUN: llc < %s -march=ptx -mattr=ptx14 | grep ".version 1.4"
+; RUN: llc < %s -march=ptx -mattr=ptx20 | grep ".version 2.0"
+; RUN: llc < %s -march=ptx -mattr=ptx21 | grep ".version 2.1"
+; RUN: llc < %s -march=ptx -mattr=sm20 | grep ".target sm_20"
+; RUN: llc < %s -march=ptx -mattr=sm13 | grep ".target sm_13"
 
 define ptx_device void @t1() {
 	ret void
diff --git a/test/CodeGen/PTX/st.ll b/test/CodeGen/PTX/st.ll
index 2cbacb9ee5..bbe89a1064 100644
--- a/test/CodeGen/PTX/st.ll
+++ b/test/CodeGen/PTX/st.ll
@@ -1,71 +1,382 @@
 ; RUN: llc < %s -march=ptx | FileCheck %s
 
-;CHECK: .extern .global .s32 array[];
-@array = external global [10 x i32]
+;CHECK: .extern .global .u16 array_i16[];
+@array_i16 = external global [10 x i16]
 
-;CHECK: .extern .const .s32 array_constant[];
-@array_constant = external addrspace(1) constant [10 x i32]
+;CHECK: .extern .const .u16 array_constant_i16[];
+@array_constant_i16 = external addrspace(1) constant [10 x i16]
 
-;CHECK: .extern .local .s32 array_local[];
-@array_local = external addrspace(2) global [10 x i32]
+;CHECK: .extern .local .u16 array_local_i16[];
+@array_local_i16 = external addrspace(2) global [10 x i16]
 
-;CHECK: .extern .shared .s32 array_shared[];
-@array_shared = external addrspace(4) global [10 x i32]
+;CHECK: .extern .shared .u16 array_shared_i16[];
+@array_shared_i16 = external addrspace(4) global [10 x i16]
 
-define ptx_device void @t1(i32* %p, i32 %x) {
+;CHECK: .extern .global .u32 array_i32[];
+@array_i32 = external global [10 x i32]
+
+;CHECK: .extern .const .u32 array_constant_i32[];
+@array_constant_i32 = external addrspace(1) constant [10 x i32]
+
+;CHECK: .extern .local .u32 array_local_i32[];
+@array_local_i32 = external addrspace(2) global [10 x i32]
+
+;CHECK: .extern .shared .u32 array_shared_i32[];
+@array_shared_i32 = external addrspace(4) global [10 x i32]
+
+;CHECK: .extern .global .u64 array_i64[];
+@array_i64 = external global [10 x i64]
+
+;CHECK: .extern .const .u64 array_constant_i64[];
+@array_constant_i64 = external addrspace(1) constant [10 x i64]
+
+;CHECK: .extern .local .u64 array_local_i64[];
+@array_local_i64 = external addrspace(2) global [10 x i64]
+
+;CHECK: .extern .shared .u64 array_shared_i64[];
+@array_shared_i64 = external addrspace(4) global [10 x i64]
+
+;CHECK: .extern .global .f32 array_float[];
+@array_float = external global [10 x float]
+
+;CHECK: .extern .const .f32 array_constant_float[];
+@array_constant_float = external addrspace(1) constant [10 x float]
+
+;CHECK: .extern .local .f32 array_local_float[];
+@array_local_float = external addrspace(2) global [10 x float]
+
+;CHECK: .extern .shared .f32 array_shared_float[];
+@array_shared_float = external addrspace(4) global [10 x float]
+
+;CHECK: .extern .global .f64 array_double[];
+@array_double = external global [10 x double]
+
+;CHECK: .extern .const .f64 array_constant_double[];
+@array_constant_double = external addrspace(1) constant [10 x double]
+
+;CHECK: .extern .local .f64 array_local_double[];
+@array_local_double = external addrspace(2) global [10 x double]
+
+;CHECK: .extern .shared .f64 array_shared_double[];
+@array_shared_double = external addrspace(4) global [10 x double]
+
+
+define ptx_device void @t1_u16(i16* %p, i16 %x) {
 entry:
-;CHECK: st.global.s32 [r1], r2;
+;CHECK: st.global.u16 [r1], rh1;
+;CHECK-NEXT: ret;
+  store i16 %x, i16* %p
+  ret void
+}
+
+define ptx_device void @t1_u32(i32* %p, i32 %x) {
+entry:
+;CHECK: st.global.u32 [r1], r2;
+;CHECK-NEXT: ret;
   store i32 %x, i32* %p
   ret void
 }
 
-define ptx_device void @t2(i32* %p, i32 %x) {
+define ptx_device void @t1_u64(i64* %p, i64 %x) {
+entry:
+;CHECK: st.global.u64 [r1], rd1;
+;CHECK-NEXT: ret;
+  store i64 %x, i64* %p
+  ret void
+}
+
+define ptx_device void @t1_f32(float* %p, float %x) {
+entry:
+;CHECK: st.global.f32 [r1], f1;
+;CHECK-NEXT: ret;
+  store float %x, float* %p
+  ret void
+}
+
+define ptx_device void @t1_f64(double* %p, double %x) {
+entry:
+;CHECK: st.global.f64 [r1], fd1;
+;CHECK-NEXT: ret;
+  store double %x, double* %p
+  ret void
+}
+
+define ptx_device void @t2_u16(i16* %p, i16 %x) {
 entry:
-;CHECK: st.global.s32 [r1+4], r2;
+;CHECK: st.global.u16 [r1+2], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr i16* %p, i32 1
+  store i16 %x, i16* %i
+  ret void
+}
+
+define ptx_device void @t2_u32(i32* %p, i32 %x) {
+entry:
+;CHECK: st.global.u32 [r1+4], r2;
+;CHECK-NEXT: ret;
   %i = getelementptr i32* %p, i32 1
   store i32 %x, i32* %i
   ret void
 }
 
-define ptx_device void @t3(i32* %p, i32 %q, i32 %x) {
-;CHECK: .reg .s32 r0;
+define ptx_device void @t2_u64(i64* %p, i64 %x) {
+entry:
+;CHECK: st.global.u64 [r1+8], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr i64* %p, i32 1
+  store i64 %x, i64* %i
+  ret void
+}
+
+define ptx_device void @t2_f32(float* %p, float %x) {
+entry:
+;CHECK: st.global.f32 [r1+4], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr float* %p, i32 1
+  store float %x, float* %i
+  ret void
+}
+
+define ptx_device void @t2_f64(double* %p, double %x) {
+entry:
+;CHECK: st.global.f64 [r1+8], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr double* %p, i32 1
+  store double %x, double* %i
+  ret void
+}
+
+define ptx_device void @t3_u16(i16* %p, i32 %q, i16 %x) {
+entry:
+;CHECK: shl.b32 r0, r2, 1;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: st.global.u16 [r0], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr i16* %p, i32 %q
+  store i16 %x, i16* %i
+  ret void
+}
+
+define ptx_device void @t3_u32(i32* %p, i32 %q, i32 %x) {
 entry:
 ;CHECK: shl.b32 r0, r2, 2;
-;CHECK: add.s32 r0, r1, r0;
-;CHECK: st.global.s32 [r0], r3;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: st.global.u32 [r0], r3;
+;CHECK-NEXT: ret;
   %i = getelementptr i32* %p, i32 %q
   store i32 %x, i32* %i
   ret void
 }
 
-define ptx_device void @t4_global(i32 %x) {
+define ptx_device void @t3_u64(i64* %p, i32 %q, i64 %x) {
+entry:
+;CHECK: shl.b32 r0, r2, 3;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: st.global.u64 [r0], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr i64* %p, i32 %q
+  store i64 %x, i64* %i
+  ret void
+}
+
+define ptx_device void @t3_f32(float* %p, i32 %q, float %x) {
+entry:
+;CHECK: shl.b32 r0, r2, 2;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: st.global.f32 [r0], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr float* %p, i32 %q
+  store float %x, float* %i
+  ret void
+}
+
+define ptx_device void @t3_f64(double* %p, i32 %q, double %x) {
+entry:
+;CHECK: shl.b32 r0, r2, 3;
+;CHECK-NEXT: add.u32 r0, r1, r0;
+;CHECK-NEXT: st.global.f64 [r0], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr double* %p, i32 %q
+  store double %x, double* %i
+  ret void
+}
+
+define ptx_device void @t4_global_u16(i16 %x) {
+entry:
+;CHECK: st.global.u16 [array_i16], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16]* @array_i16, i16 0, i16 0
+  store i16 %x, i16* %i
+  ret void
+}
+
+define ptx_device void @t4_global_u32(i32 %x) {
 entry:
-;CHECK: st.global.s32 [array], r1;
-  %i = getelementptr [10 x i32]* @array, i32 0, i32 0
+;CHECK: st.global.u32 [array_i32], r1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32]* @array_i32, i32 0, i32 0
   store i32 %x, i32* %i
   ret void
 }
 
-define ptx_device void @t4_local(i32 %x) {
+define ptx_device void @t4_global_u64(i64 %x) {
+entry:
+;CHECK: st.global.u64 [array_i64], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64]* @array_i64, i32 0, i32 0
+  store i64 %x, i64* %i
+  ret void
+}
+
+define ptx_device void @t4_global_f32(float %x) {
+entry:
+;CHECK: st.global.f32 [array_float], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float]* @array_float, i32 0, i32 0
+  store float %x, float* %i
+  ret void
+}
+
+define ptx_device void @t4_global_f64(double %x) {
+entry:
+;CHECK: st.global.f64 [array_double], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double]* @array_double, i32 0, i32 0
+  store double %x, double* %i
+  ret void
+}
+
+define ptx_device void @t4_local_u16(i16 %x) {
 entry:
-;CHECK: st.local.s32 [array_local], r1;
-  %i = getelementptr [10 x i32] addrspace(2)* @array_local, i32 0, i32 0
+;CHECK: st.local.u16 [array_local_i16], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16] addrspace(2)* @array_local_i16, i32 0, i32 0
+  store i16 %x, i16 addrspace(2)* %i
+  ret void
+}
+
+define ptx_device void @t4_local_u32(i32 %x) {
+entry:
+;CHECK: st.local.u32 [array_local_i32], r1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32] addrspace(2)* @array_local_i32, i32 0, i32 0
   store i32 %x, i32 addrspace(2)* %i
   ret void
 }
 
-define ptx_device void @t4_shared(i32 %x) {
+define ptx_device void @t4_local_u64(i64 %x) {
+entry:
+;CHECK: st.local.u64 [array_local_i64], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64] addrspace(2)* @array_local_i64, i32 0, i32 0
+  store i64 %x, i64 addrspace(2)* %i
+  ret void
+}
+
+define ptx_device void @t4_local_f32(float %x) {
+entry:
+;CHECK: st.local.f32 [array_local_float], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float] addrspace(2)* @array_local_float, i32 0, i32 0
+  store float %x, float addrspace(2)* %i
+  ret void
+}
+
+define ptx_device void @t4_local_f64(double %x) {
+entry:
+;CHECK: st.local.f64 [array_local_double], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double] addrspace(2)* @array_local_double, i32 0, i32 0
+  store double %x, double addrspace(2)* %i
+  ret void
+}
+
+define ptx_device void @t4_shared_u16(i16 %x) {
 entry:
-;CHECK: st.shared.s32 [array_shared], r1;
-  %i = getelementptr [10 x i32] addrspace(4)* @array_shared, i32 0, i32 0
+;CHECK: st.shared.u16 [array_shared_i16], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16] addrspace(4)* @array_shared_i16, i32 0, i32 0
+  store i16 %x, i16 addrspace(4)* %i
+  ret void
+}
+
+define ptx_device void @t4_shared_u32(i32 %x) {
+entry:
+;CHECK: st.shared.u32 [array_shared_i32], r1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32] addrspace(4)* @array_shared_i32, i32 0, i32 0
   store i32 %x, i32 addrspace(4)* %i
   ret void
 }
 
-define ptx_device void @t5(i32 %x) {
+define ptx_device void @t4_shared_u64(i64 %x) {
+entry:
+;CHECK: st.shared.u64 [array_shared_i64], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64] addrspace(4)* @array_shared_i64, i32 0, i32 0
+  store i64 %x, i64 addrspace(4)* %i
+  ret void
+}
+
+define ptx_device void @t4_shared_f32(float %x) {
+entry:
+;CHECK: st.shared.f32 [array_shared_float], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float] addrspace(4)* @array_shared_float, i32 0, i32 0
+  store float %x, float addrspace(4)* %i
+  ret void
+}
+
+define ptx_device void @t4_shared_f64(double %x) {
+entry:
+;CHECK: st.shared.f64 [array_shared_double], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double] addrspace(4)* @array_shared_double, i32 0, i32 0
+  store double %x, double addrspace(4)* %i
+  ret void
+}
+
+define ptx_device void @t5_u16(i16 %x) {
 entry:
-;CHECK: st.global.s32 [array+4], r1;
-  %i = getelementptr [10 x i32]* @array, i32 0, i32 1
+;CHECK: st.global.u16 [array_i16+2], rh1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i16]* @array_i16, i32 0, i32 1
+  store i16 %x, i16* %i
+  ret void
+}
+
+define ptx_device void @t5_u32(i32 %x) {
+entry:
+;CHECK: st.global.u32 [array_i32+4], r1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i32]* @array_i32, i32 0, i32 1
   store i32 %x, i32* %i
   ret void
 }
+
+define ptx_device void @t5_u64(i64 %x) {
+entry:
+;CHECK: st.global.u64 [array_i64+8], rd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x i64]* @array_i64, i32 0, i32 1
+  store i64 %x, i64* %i
+  ret void
+}
+
+define ptx_device void @t5_f32(float %x) {
+entry:
+;CHECK: st.global.f32 [array_float+4], f1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x float]* @array_float, i32 0, i32 1
+  store float %x, float* %i
+  ret void
+}
+
+define ptx_device void @t5_f64(double %x) {
+entry:
+;CHECK: st.global.f64 [array_double+8], fd1;
+;CHECK-NEXT: ret;
+  %i = getelementptr [10 x double]* @array_double, i32 0, i32 1
+  store double %x, double* %i
+  ret void
+}
diff --git a/test/CodeGen/PTX/st_float.ll b/test/CodeGen/PTX/st_float.ll
deleted file mode 100644
index f0e00105f9..0000000000
--- a/test/CodeGen/PTX/st_float.ll
+++ /dev/null
@@ -1,78 +0,0 @@
-; RUN: llc < %s -march=ptx | FileCheck %s
-
-;CHECK: .extern .global .f32 array[];
-@array = external global [10 x float]
-
-;CHECK: .extern .const .f32 array_constant[];
-@array_constant = external addrspace(1) constant [10 x float]
-
-;CHECK: .extern .local .f32 array_local[];
-@array_local = external addrspace(2) global [10 x float]
-
-;CHECK: .extern .shared .f32 array_shared[];
-@array_shared = external addrspace(4) global [10 x float]
-
-define ptx_device void @t1(float* %p, float %x) {
-entry:
-;CHECK: st.global.f32 [r1], f1;
-;CHECK-NEXT: ret;
-  store float %x, float* %p
-  ret void
-}
-
-define ptx_device void @t2(float* %p, float %x) {
-entry:
-;CHECK: st.global.f32 [r1+4], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr float* %p, i32 1
-  store float %x, float* %i
-  ret void
-}
-
-define ptx_device void @t3(float* %p, i32 %q, float %x) {
-;CHECK: .reg .s32 r0;
-entry:
-;CHECK: shl.b32 r0, r2, 2;
-;CHECK-NEXT: add.s32 r0, r1, r0;
-;CHECK-NEXT: st.global.f32 [r0], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr float* %p, i32 %q
-  store float %x, float* %i
-  ret void
-}
-
-define ptx_device void @t4_global(float %x) {
-entry:
-;CHECK: st.global.f32 [array], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float]* @array, i32 0, i32 0
-  store float %x, float* %i
-  ret void
-}
-
-define ptx_device void @t4_local(float %x) {
-entry:
-;CHECK: st.local.f32 [array_local], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float] addrspace(2)* @array_local, i32 0, i32 0
-  store float %x, float addrspace(2)* %i
-  ret void
-}
-
-define ptx_device void @t4_shared(float %x) {
-entry:
-;CHECK: st.shared.f32 [array_shared], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float] addrspace(4)* @array_shared, i32 0, i32 0
-  store float %x, float addrspace(4)* %i
-  ret void
-}
-
-define ptx_device void @t5(float %x) {
-entry:
-;CHECK: st.global.f32 [array+4], f1;
-;CHECK-NEXT: ret;
-  %i = getelementptr [10 x float]* @array, i32 0, i32 1
-  store float %x, float* %i
-  ret void
-}
diff --git a/test/CodeGen/PTX/sub.ll b/test/CodeGen/PTX/sub.ll
index e11decaf5c..4810e4fc05 100644
--- a/test/CodeGen/PTX/sub.ll
+++ b/test/CodeGen/PTX/sub.ll
@@ -1,29 +1,71 @@
 ; RUN: llc < %s -march=ptx | FileCheck %s
 
-define ptx_device i32 @t1(i32 %x, i32 %y) {
-;CHECK: sub.s32 r0, r1, r2;
+define ptx_device i16 @t1_u16(i16 %x, i16 %y) {
+; CHECK: sub.u16 rh0, rh1, rh2;
+; CHECK-NEXT: ret;
+	%z = sub i16 %x, %y
+	ret i16 %z
+}
+
+define ptx_device i32 @t1_u32(i32 %x, i32 %y) {
+; CHECK: sub.u32 r0, r1, r2;
+; CHECK-NEXT: ret;
 	%z = sub i32 %x, %y
-;CHECK: ret;
 	ret i32 %z
 }
 
-define ptx_device i32 @t2(i32 %x) {
-;CHECK: add.s32 r0, r1, -1;
-	%z = sub i32 %x, 1
-;CHECK: ret;
-	ret i32 %z
+define ptx_device i64 @t1_u64(i64 %x, i64 %y) {
+; CHECK: sub.u64 rd0, rd1, rd2;
+; CHECK-NEXT: ret;
+	%z = sub i64 %x, %y
+	ret i64 %z
 }
 
-define ptx_device float @t3(float %x, float %y) {
+define ptx_device float @t1_f32(float %x, float %y) {
 ; CHECK: sub.f32 f0, f1, f2
 ; CHECK-NEXT: ret;
   %z = fsub float %x, %y
   ret float %z
 }
 
-define ptx_device float @t4(float %x) {
+define ptx_device double @t1_f64(double %x, double %y) {
+; CHECK: sub.f64 fd0, fd1, fd2
+; CHECK-NEXT: ret;
+  %z = fsub double %x, %y
+  ret double %z
+}
+
+define ptx_device i16 @t2_u16(i16 %x) {
+; CHECK: add.u16 rh0, rh1, -1;
+; CHECK-NEXT: ret;
+	%z = sub i16 %x, 1
+	ret i16 %z
+}
+
+define ptx_device i32 @t2_u32(i32 %x) {
+; CHECK: add.u32 r0, r1, -1;
+; CHECK-NEXT: ret;
+	%z = sub i32 %x, 1
+	ret i32 %z
+}
+
+define ptx_device i64 @t2_u64(i64 %x) {
+; CHECK: add.u64 rd0, rd1, -1;
+; CHECK-NEXT: ret;
+	%z = sub i64 %x, 1
+	ret i64 %z
+}
+
+define ptx_device float @t2_f32(float %x) {
 ; CHECK: add.f32 f0, f1, 0FBF800000;
 ; CHECK-NEXT: ret;
   %z = fsub float %x, 1.0
   ret float %z
 }
+
+define ptx_device double @t2_f64(double %x) {
+; CHECK: add.f64 fd0, fd1, 0DBFF0000000000000;
+; CHECK-NEXT: ret;
+  %z = fsub double %x, 1.0
+  ret double %z
+}