[X86] Add additional test coverage for #160111 (#162838)

Author: RKSimon

llvm/test/CodeGen/X86/fp-int-fp-cvt.ll

@@ -0,0 +1,240 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64    | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX512,AVX512-VL
+; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v4 -mattr=-avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512-NOVL
+
+;
+; fptosi -> sitofp
+;
+
+define double @scvtf64_i32(double %a0) {
+; SSE-LABEL: scvtf64_i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttpd2dq %xmm0, %xmm0
+; SSE-NEXT:    cvtdq2pd %xmm0, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: scvtf64_i32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vcvttpd2dq %xmm0, %xmm0
+; AVX-NEXT:    vcvtdq2pd %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %ii = fptosi double %a0 to i32
+  %ff = sitofp i32 %ii to double
+  ret double %ff
+}
+
+define double @scvtf64_i64(double %a0) {
+; SSE-LABEL: scvtf64_i64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttsd2si %xmm0, %rax
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2sd %rax, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: scvtf64_i64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vcvttsd2si %xmm0, %rax
+; AVX-NEXT:    vcvtsi2sd %rax, %xmm15, %xmm0
+; AVX-NEXT:    retq
+  %ii = fptosi double %a0 to i64
+  %ff = sitofp i64 %ii to double
+  ret double %ff
+}
+
+define float @scvtf32_i32(float %a0) {
+; SSE-LABEL: scvtf32_i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttps2dq %xmm0, %xmm0
+; SSE-NEXT:    cvtdq2ps %xmm0, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: scvtf32_i32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vcvttps2dq %xmm0, %xmm0
+; AVX-NEXT:    vcvtdq2ps %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %ii = fptosi float %a0 to i32
+  %ff = sitofp i32 %ii to float
+  ret float %ff
+}
+
+define float @scvtf32_i64(float %a0) {
+; SSE-LABEL: scvtf32_i64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttss2si %xmm0, %rax
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2ss %rax, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: scvtf32_i64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vcvttss2si %xmm0, %rax
+; AVX-NEXT:    vcvtsi2ss %rax, %xmm15, %xmm0
+; AVX-NEXT:    retq
+  %ii = fptosi float %a0 to i64
+  %ff = sitofp i64 %ii to float
+  ret float %ff
+}
+
+;
+; fptoui -> uitofp
+;
+
+define double @ucvtf64_i32(double %a0) {
+; SSE-LABEL: ucvtf64_i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttsd2si %xmm0, %rax
+; SSE-NEXT:    movl %eax, %eax
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2sd %rax, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX2-LABEL: ucvtf64_i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vcvttsd2si %xmm0, %rax
+; AVX2-NEXT:    movl %eax, %eax
+; AVX2-NEXT:    vcvtsi2sd %rax, %xmm15, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: ucvtf64_i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvttsd2usi %xmm0, %eax
+; AVX512-NEXT:    vcvtusi2sd %eax, %xmm15, %xmm0
+; AVX512-NEXT:    retq
+  %ii = fptoui double %a0 to i32
+  %ff = uitofp i32 %ii to double
+  ret double %ff
+}
+
+define double @ucvtf64_i64(double %a0) {
+; SSE-LABEL: ucvtf64_i64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttsd2si %xmm0, %rax
+; SSE-NEXT:    movq %rax, %rcx
+; SSE-NEXT:    subsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    cvttsd2si %xmm0, %rdx
+; SSE-NEXT:    sarq $63, %rcx
+; SSE-NEXT:    andq %rcx, %rdx
+; SSE-NEXT:    orq %rax, %rdx
+; SSE-NEXT:    movq %rdx, %xmm1
+; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; SSE-NEXT:    subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE-NEXT:    movapd %xmm1, %xmm0
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; SSE-NEXT:    addsd %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX2-LABEL: ucvtf64_i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vcvttsd2si %xmm0, %rax
+; AVX2-NEXT:    movq %rax, %rcx
+; AVX2-NEXT:    vsubsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    sarq $63, %rcx
+; AVX2-NEXT:    vcvttsd2si %xmm0, %rdx
+; AVX2-NEXT:    andq %rcx, %rdx
+; AVX2-NEXT:    orq %rax, %rdx
+; AVX2-NEXT:    vmovq %rdx, %xmm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; AVX2-NEXT:    vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX2-NEXT:    vaddsd %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: ucvtf64_i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvttsd2usi %xmm0, %rax
+; AVX512-NEXT:    vcvtusi2sd %rax, %xmm15, %xmm0
+; AVX512-NEXT:    retq
+  %ii = fptoui double %a0 to i64
+  %ff = uitofp i64 %ii to double
+  ret double %ff
+}
+
+define float @ucvtf32_i32(float %a0) {
+; SSE-LABEL: ucvtf32_i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttss2si %xmm0, %rax
+; SSE-NEXT:    movl %eax, %eax
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2ss %rax, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX2-LABEL: ucvtf32_i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vcvttss2si %xmm0, %rax
+; AVX2-NEXT:    movl %eax, %eax
+; AVX2-NEXT:    vcvtsi2ss %rax, %xmm15, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: ucvtf32_i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvttss2usi %xmm0, %eax
+; AVX512-NEXT:    vcvtusi2ss %eax, %xmm15, %xmm0
+; AVX512-NEXT:    retq
+  %ii = fptoui float %a0 to i32
+  %ff = uitofp i32 %ii to float
+  ret float %ff
+}
+
+define float @ucvtf32_i64(float %a0) {
+; SSE-LABEL: ucvtf32_i64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    cvttss2si %xmm0, %rcx
+; SSE-NEXT:    movq %rcx, %rdx
+; SSE-NEXT:    sarq $63, %rdx
+; SSE-NEXT:    subss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    cvttss2si %xmm0, %rax
+; SSE-NEXT:    andq %rdx, %rax
+; SSE-NEXT:    orq %rcx, %rax
+; SSE-NEXT:    js .LBB7_1
+; SSE-NEXT:  # %bb.2:
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2ss %rax, %xmm0
+; SSE-NEXT:    retq
+; SSE-NEXT:  .LBB7_1:
+; SSE-NEXT:    movq %rax, %rcx
+; SSE-NEXT:    shrq %rcx
+; SSE-NEXT:    andl $1, %eax
+; SSE-NEXT:    orq %rcx, %rax
+; SSE-NEXT:    xorps %xmm0, %xmm0
+; SSE-NEXT:    cvtsi2ss %rax, %xmm0
+; SSE-NEXT:    addss %xmm0, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX2-LABEL: ucvtf32_i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vcvttss2si %xmm0, %rcx
+; AVX2-NEXT:    movq %rcx, %rdx
+; AVX2-NEXT:    sarq $63, %rdx
+; AVX2-NEXT:    vsubss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vcvttss2si %xmm0, %rax
+; AVX2-NEXT:    andq %rdx, %rax
+; AVX2-NEXT:    orq %rcx, %rax
+; AVX2-NEXT:    js .LBB7_1
+; AVX2-NEXT:  # %bb.2:
+; AVX2-NEXT:    vcvtsi2ss %rax, %xmm15, %xmm0
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB7_1:
+; AVX2-NEXT:    movq %rax, %rcx
+; AVX2-NEXT:    shrq %rcx
+; AVX2-NEXT:    andl $1, %eax
+; AVX2-NEXT:    orq %rcx, %rax
+; AVX2-NEXT:    vcvtsi2ss %rax, %xmm15, %xmm0
+; AVX2-NEXT:    vaddss %xmm0, %xmm0, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: ucvtf32_i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vcvttss2usi %xmm0, %rax
+; AVX512-NEXT:    vcvtusi2ss %rax, %xmm15, %xmm0
+; AVX512-NEXT:    retq
+  %ii = fptoui float %a0 to i64
+  %ff = uitofp i64 %ii to float
+  ret float %ff
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; AVX512-NOVL: {{.*}}
+; AVX512-VL: {{.*}}