diff options
Diffstat (limited to 'lib/Target/X86/README.txt')
| -rw-r--r-- | lib/Target/X86/README.txt | 21 |
1 files changed, 16 insertions, 5 deletions
diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt index be37acfa8a..b239dc3edd 100644 --- a/lib/Target/X86/README.txt +++ b/lib/Target/X86/README.txt @@ -1601,13 +1601,13 @@ This would result in smaller code and more efficient microops. In SSE mode, we turn abs and neg into a load from the constant pool plus a xor or and instruction, for example: - xorpd LCPI2_0-"L2$pb"(%esi), %xmm2 + xorpd LCPI1_0, %xmm2 However, if xmm2 gets spilled, we end up with really ugly code like this: - %xmm2 = reload [mem] - xorpd LCPI2_0-"L2$pb"(%esi), %xmm2 - store %xmm2 -> [mem] + movsd (%esp), %xmm0 + xorpd LCPI1_0, %xmm0 + movsd %xmm0, (%esp) Since we 'know' that this is a 'neg', we can actually "fold" the spill into the neg/abs instruction, turning it into an *integer* operation, like this: @@ -1615,6 +1615,17 @@ the neg/abs instruction, turning it into an *integer* operation, like this: xorl 2147483648, [mem+4] ## 2147483648 = (1 << 31) you could also use xorb, but xorl is less likely to lead to a partial register -stall. +stall. Here is a contrived testcase: + +double a, b, c; +void test(double *P) { + double X = *P; + a = X; + bar(); + X = -X; + b = X; + bar(); + c = X; +} //===---------------------------------------------------------------------===// |