summaryrefslogtreecommitdiffstats
path: root/llvm/test
diff options
context:
space:
mode:
authorAhmed Bougacha <ahmed.bougacha@gmail.com>2017-06-02 20:02:59 +0000
committerAhmed Bougacha <ahmed.bougacha@gmail.com>2017-06-02 20:02:59 +0000
commit018a68f9e4e5c77a14cce36e334e3323a61e0aaf (patch)
tree7aef4cb2bdf5c55f5bc693edaa5e2aad2feab97b /llvm/test
parent4bedb5fd00eb271d62571eadcd6042ea30538c9e (diff)
downloadbcm5719-llvm-018a68f9e4e5c77a14cce36e334e3323a61e0aaf.tar.gz
bcm5719-llvm-018a68f9e4e5c77a14cce36e334e3323a61e0aaf.zip
[X86] Correctly broadcast NaN-like integers as float on AVX.
Since r288804, we try to lower build_vectors on AVX using broadcasts of float/double. However, when we broadcast integer values that happen to have a NaN float bitpattern, we lose the NaN payload, thereby changing the integer value being broadcast. This is caused by ConstantFP::get, to which we pass the splat i32 as a float (by bitcasting it using bitsToFloat). ConstantFP::get takes a double parameter, so we end up lossily converting a single-precision NaN to double-precision. Instead, avoid any kinds of conversions by directly building an APFloat from the splatted APInt. Note that this also fixes another piece of code (broadcast of subvectors), that currently isn't susceptible to the same problem. Also note that we could really just use APInt and ConstantInt throughout: the constant pool type doesn't matter much. Still, for consistency, use the appropriate type. llvm-svn: 304590
Diffstat (limited to 'llvm/test')
-rw-r--r--llvm/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll32
1 files changed, 32 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll b/llvm/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
index b13965a30ed..bbe31c5c2ac 100644
--- a/llvm/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
+++ b/llvm/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
@@ -1203,3 +1203,35 @@ define <8 x double> @f8xf64_f256(<8 x double> %a) {
ret <8 x double> %res2
}
+
+
+; ALL: .LCPI38
+; ALL-NEXT: .long 4290379776 # 0xffba0000
+
+; AVX: .LCPI38
+; AVX-NEXT: .long 4290379776 # float NaN
+
+define <8 x i16> @f8xi16_i32_NaN(<8 x i16> %a) {
+; ALL32-LABEL: f8xi16_i32_NaN:
+; ALL32: # BB#0:
+; ALL32-NEXT: vpbroadcastd {{\.LCPI.*}}, %xmm1
+; ALL32-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; ALL32-NEXT: vpand %xmm1, %xmm0, %xmm0
+; ALL32-NEXT: retl
+;
+; ALL64-LABEL: f8xi16_i32_NaN:
+; ALL64: # BB#0:
+; ALL64-NEXT: vpbroadcastd {{.*}}(%rip), %xmm1
+; ALL64-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; ALL64-NEXT: vpand %xmm1, %xmm0, %xmm0
+; ALL64-NEXT: retq
+;
+; AVX-LABEL: f8xi16_i32_NaN:
+; AVX: # BB#0:
+; AVX-NEXT: vbroadcastss {{\.LCPI.*}}, %xmm1
+; AVX-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
+ %res1 = add <8 x i16> <i16 0, i16 -70, i16 0, i16 -70, i16 0, i16 -70, i16 0, i16 -70>, %a
+ %res2 = and <8 x i16> <i16 0, i16 -70, i16 0, i16 -70, i16 0, i16 -70, i16 0, i16 -70>, %res1
+ ret <8 x i16> %res2
+}
OpenPOWER on IntegriCloud