[x86] add tests for select-of-fp-constants; NFC

There are many options here depending on subtarget,
but we are uniformly relying on a transform that was 
driven by performance for a 32-bit SSE2 target in 2009.

Note: The same motivation was apparently used to do this 
transform for *all* targets, so non-x86 may want to look
at this too.

llvm-svn: 347525

2 files changed, 96 insertions, 18 deletions

diff --git a/llvm/test/CodeGen/X86/2009-03-07-FPConstSelect.ll b/llvm/test/CodeGen/X86/2009-03-07-FPConstSelect.ll
deleted file mode 100644
index d38e8b2d638..00000000000
--- a/llvm/test/CodeGen/X86/2009-03-07-FPConstSelect.ll
+++ /dev/null
@@ -1,18 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i386-- | FileCheck %s
-
-; This should do a single load into the fp stack for the return, not diddle with xmm registers.
-
-define float @f(i32 %x) nounwind readnone {
-; CHECK-LABEL: f:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    cmpl $0, {{[0-9]+}}(%esp)
-; CHECK-NEXT:    sete %al
-; CHECK-NEXT:    flds {{\.LCPI.*}}(,%eax,4)
-; CHECK-NEXT:    retl
-	%c = icmp eq i32 %x, 0
-	%r = select i1 %c, float 42.0, float 23.0
-	ret float %r
-}
-
diff --git a/llvm/test/CodeGen/X86/select-of-fp-constants.ll b/llvm/test/CodeGen/X86/select-of-fp-constants.ll
new file mode 100644
index 00000000000..84b0bc13441
--- /dev/null
+++ b/llvm/test/CodeGen/X86/select-of-fp-constants.ll
@@ -0,0 +1,96 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i386--   -mattr=sse2     | FileCheck %s --check-prefixes=X32,X32_SSE,X32_SSE2
+; RUN: llc < %s -mtriple=i386--   -mattr=sse4.1   | FileCheck %s --check-prefixes=X32,X32_SSE,X32_SSE4
+; RUN: llc < %s -mtriple=i386--   -mattr=avx2     | FileCheck %s --check-prefixes=X32,X32_AVX,X32_AVX2
+; RUN: llc < %s -mtriple=i386--   -mattr=avx512f  | FileCheck %s --check-prefixes=X32,X32_AVX,X32_AVX512F
+; RUN: llc < %s -mtriple=x86_64-- -mattr=sse2     | FileCheck %s --check-prefixes=X64,X64_SSE,X64_SSE2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=sse4.1   | FileCheck %s --check-prefixes=X64,X64_SSE,X64_SSE4
+; RUN: llc < %s -mtriple=x86_64-- -mattr=avx2     | FileCheck %s --check-prefixes=X64,X64_AVX,X64_AVX2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=avx512f  | FileCheck %s --check-prefixes=X64,X64_AVX,X64_AVX512F
+
+; This should do a single load into the fp stack for the return, not diddle with xmm registers.
+
+define float @icmp_select_fp_constants(i32 %x) nounwind readnone {
+; X32-LABEL: icmp_select_fp_constants:
+; X32:       # %bb.0:
+; X32-NEXT:    xorl %eax, %eax
+; X32-NEXT:    cmpl $0, {{[0-9]+}}(%esp)
+; X32-NEXT:    sete %al
+; X32-NEXT:    flds {{\.LCPI.*}}(,%eax,4)
+; X32-NEXT:    retl
+;
+; X64_SSE-LABEL: icmp_select_fp_constants:
+; X64_SSE:       # %bb.0:
+; X64_SSE-NEXT:    xorl %eax, %eax
+; X64_SSE-NEXT:    testl %edi, %edi
+; X64_SSE-NEXT:    sete %al
+; X64_SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64_SSE-NEXT:    retq
+;
+; X64_AVX-LABEL: icmp_select_fp_constants:
+; X64_AVX:       # %bb.0:
+; X64_AVX-NEXT:    xorl %eax, %eax
+; X64_AVX-NEXT:    testl %edi, %edi
+; X64_AVX-NEXT:    sete %al
+; X64_AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64_AVX-NEXT:    retq
+	%c = icmp eq i32 %x, 0
+	%r = select i1 %c, float 42.0, float 23.0
+	ret float %r
+}
+
+define float @fcmp_select_fp_constants(float %x) nounwind readnone {
+; X32_SSE-LABEL: fcmp_select_fp_constants:
+; X32_SSE:       # %bb.0:
+; X32_SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X32_SSE-NEXT:    cmpneqss {{[0-9]+}}(%esp), %xmm0
+; X32_SSE-NEXT:    movd %xmm0, %eax
+; X32_SSE-NEXT:    andl $1, %eax
+; X32_SSE-NEXT:    flds {{\.LCPI.*}}(,%eax,4)
+; X32_SSE-NEXT:    retl
+;
+; X32_AVX2-LABEL: fcmp_select_fp_constants:
+; X32_AVX2:       # %bb.0:
+; X32_AVX2-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X32_AVX2-NEXT:    vcmpneqss {{[0-9]+}}(%esp), %xmm0, %xmm0
+; X32_AVX2-NEXT:    vmovd %xmm0, %eax
+; X32_AVX2-NEXT:    andl $1, %eax
+; X32_AVX2-NEXT:    flds {{\.LCPI.*}}(,%eax,4)
+; X32_AVX2-NEXT:    retl
+;
+; X32_AVX512F-LABEL: fcmp_select_fp_constants:
+; X32_AVX512F:       # %bb.0:
+; X32_AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X32_AVX512F-NEXT:    vcmpneqss {{[0-9]+}}(%esp), %xmm0, %k0
+; X32_AVX512F-NEXT:    kmovw %k0, %eax
+; X32_AVX512F-NEXT:    flds {{\.LCPI.*}}(,%eax,4)
+; X32_AVX512F-NEXT:    retl
+;
+; X64_SSE-LABEL: fcmp_select_fp_constants:
+; X64_SSE:       # %bb.0:
+; X64_SSE-NEXT:    cmpneqss {{.*}}(%rip), %xmm0
+; X64_SSE-NEXT:    movd %xmm0, %eax
+; X64_SSE-NEXT:    andl $1, %eax
+; X64_SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64_SSE-NEXT:    retq
+;
+; X64_AVX2-LABEL: fcmp_select_fp_constants:
+; X64_AVX2:       # %bb.0:
+; X64_AVX2-NEXT:    vcmpneqss {{.*}}(%rip), %xmm0, %xmm0
+; X64_AVX2-NEXT:    vmovd %xmm0, %eax
+; X64_AVX2-NEXT:    andl $1, %eax
+; X64_AVX2-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64_AVX2-NEXT:    retq
+;
+; X64_AVX512F-LABEL: fcmp_select_fp_constants:
+; X64_AVX512F:       # %bb.0:
+; X64_AVX512F-NEXT:    vcmpneqss {{.*}}(%rip), %xmm0, %k0
+; X64_AVX512F-NEXT:    kmovw %k0, %eax
+; X64_AVX512F-NEXT:    movzwl %ax, %eax
+; X64_AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64_AVX512F-NEXT:    retq
+ %c = fcmp une float %x, -4.0
+ %r = select i1 %c, float 42.0, float 23.0
+ ret float %r
+}
+