Mercurial > hg > graal-jvmci-8
comparison src/cpu/x86/vm/x86_64.ad @ 1914:ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
Summary: Use hardware DIV instruction for long division by constant when it is faster than code with multiply.
Reviewed-by: never
| author | kvn |
|---|---|
| date | Tue, 02 Nov 2010 09:00:37 -0700 |
| parents | 52e82a6bedaf |
| children | 2f644f85485d |
comparison
equal
deleted
inserted
replaced
| 1913:3b2dea75431e | 1914:ae065c367d93 |
|---|---|
| 2063 bool Matcher::is_spillable_arg(int reg) | 2063 bool Matcher::is_spillable_arg(int reg) |
| 2064 { | 2064 { |
| 2065 return can_be_java_arg(reg); | 2065 return can_be_java_arg(reg); |
| 2066 } | 2066 } |
| 2067 | 2067 |
| 2068 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) { | |
| 2069 // In 64 bit mode a code which use multiply when | |
| 2070 // devisor is constant is faster than hardware | |
| 2071 // DIV instruction (it uses MulHiL). | |
| 2072 return false; | |
| 2073 } | |
| 2074 | |
| 2068 // Register for DIVI projection of divmodI | 2075 // Register for DIVI projection of divmodI |
| 2069 RegMask Matcher::divI_proj_mask() { | 2076 RegMask Matcher::divI_proj_mask() { |
| 2070 return INT_RAX_REG_mask; | 2077 return INT_RAX_REG_mask; |
| 2071 } | 2078 } |
| 2072 | 2079 |