pre-commit tests

Author: AlexMaclean

llvm/test/CodeGen/NVPTX/i8x4-instructions.ll

@@ -1425,4 +1425,30 @@ entry:
   ret void
 }
 
+define <4 x float> @test_uitofp_v4i8(<4 x i8> %a) {
+; CHECK-LABEL: test_uitofp_v4i8(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b16 %rs<5>;
+; CHECK-NEXT:    .reg .b32 %r<10>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [test_uitofp_v4i8_param_0];
+; CHECK-NEXT:    prmt.b32 %r2, %r1, 0, 0x7773U;
+; CHECK-NEXT:    cvt.u16.u32 %rs1, %r2;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r3, %rs1;
+; CHECK-NEXT:    prmt.b32 %r4, %r1, 0, 0x7772U;
+; CHECK-NEXT:    cvt.u16.u32 %rs2, %r4;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r5, %rs2;
+; CHECK-NEXT:    prmt.b32 %r6, %r1, 0, 0x7771U;
+; CHECK-NEXT:    cvt.u16.u32 %rs3, %r6;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r7, %rs3;
+; CHECK-NEXT:    prmt.b32 %r8, %r1, 0, 0x7770U;
+; CHECK-NEXT:    cvt.u16.u32 %rs4, %r8;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r9, %rs4;
+; CHECK-NEXT:    st.param.v4.b32 [func_retval0], {%r9, %r7, %r5, %r3};
+; CHECK-NEXT:    ret;
+  %f = uitofp <4 x i8> %a to <4 x float>
+  ret <4 x float> %f
+}
+
 attributes #0 = { nounwind }

llvm/test/CodeGen/NVPTX/trunc-tofp.ll

@@ -0,0 +1,89 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mcpu=sm_80 | FileCheck %s
+; RUN: %if ptxas %{ llc < %s -mcpu=sm_80 | %ptxas-verify -arch=sm_80 %}
+
+target triple = "nvptx64-nvidia-cuda"
+
+define float @uitofp_trunc_nuw(i32 %x, i32 %y) {
+; CHECK-LABEL: uitofp_trunc_nuw(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b16 %rs<2>;
+; CHECK-NEXT:    .reg .b32 %r<5>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [uitofp_trunc_nuw_param_0];
+; CHECK-NEXT:    ld.param.b32 %r2, [uitofp_trunc_nuw_param_1];
+; CHECK-NEXT:    add.s32 %r3, %r1, %r2;
+; CHECK-NEXT:    cvt.u16.u32 %rs1, %r3;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r4, %rs1;
+; CHECK-NEXT:    st.param.b32 [func_retval0], %r4;
+; CHECK-NEXT:    ret;
+  %v = add i32 %x, %y
+  %t = trunc nuw i32 %v to i16
+  %f = uitofp i16 %t to float
+  ret float %f
+}
+
+define float @sitofp_trunc_nsw(i32 %x, i32 %y) {
+; CHECK-LABEL: sitofp_trunc_nsw(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b16 %rs<2>;
+; CHECK-NEXT:    .reg .b32 %r<5>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [sitofp_trunc_nsw_param_0];
+; CHECK-NEXT:    ld.param.b32 %r2, [sitofp_trunc_nsw_param_1];
+; CHECK-NEXT:    add.s32 %r3, %r1, %r2;
+; CHECK-NEXT:    cvt.u16.u32 %rs1, %r3;
+; CHECK-NEXT:    cvt.rn.f32.s16 %r4, %rs1;
+; CHECK-NEXT:    st.param.b32 [func_retval0], %r4;
+; CHECK-NEXT:    ret;
+  %v = add i32 %x, %y
+  %t = trunc nsw i32 %v to i16
+  %f = sitofp i16 %t to float
+  ret float %f
+}
+
+;; Cannot safely fold here because the sign of the comparison does not match the
+;; sign of the wrap flag.
+define float @uitofp_trunc_nsw(i32 %x, i32 %y) {
+; CHECK-LABEL: uitofp_trunc_nsw(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b16 %rs<2>;
+; CHECK-NEXT:    .reg .b32 %r<5>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [uitofp_trunc_nsw_param_0];
+; CHECK-NEXT:    ld.param.b32 %r2, [uitofp_trunc_nsw_param_1];
+; CHECK-NEXT:    add.s32 %r3, %r1, %r2;
+; CHECK-NEXT:    cvt.u16.u32 %rs1, %r3;
+; CHECK-NEXT:    cvt.rn.f32.u16 %r4, %rs1;
+; CHECK-NEXT:    st.param.b32 [func_retval0], %r4;
+; CHECK-NEXT:    ret;
+  %v = add i32 %x, %y
+  %t = trunc nsw i32 %v to i16
+  %f = uitofp i16 %t to float
+  ret float %f
+}
+
+;; Cannot safely fold here because the sign of the comparison does not match the
+;; sign of the wrap flag.
+define float @sitofp_trunc_nuw(i32 %x, i32 %y) {
+; CHECK-LABEL: sitofp_trunc_nuw(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b16 %rs<2>;
+; CHECK-NEXT:    .reg .b32 %r<5>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [sitofp_trunc_nuw_param_0];
+; CHECK-NEXT:    ld.param.b32 %r2, [sitofp_trunc_nuw_param_1];
+; CHECK-NEXT:    add.s32 %r3, %r1, %r2;
+; CHECK-NEXT:    cvt.u16.u32 %rs1, %r3;
+; CHECK-NEXT:    cvt.rn.f32.s16 %r4, %rs1;
+; CHECK-NEXT:    st.param.b32 [func_retval0], %r4;
+; CHECK-NEXT:    ret;
+  %v = add i32 %x, %y
+  %t = trunc nuw i32 %v to i16
+  %f = sitofp i16 %t to float
+  ret float %f
+}