Skip to content

S
swiftshader
Commit 98cc08ca by John Porto
Options

Subzero. ARM32. Strength reduce multiplications.

BUG= https://code.google.com/p/nativeclient/issues/detail?id=4076 R=stichnot@chromium.org Review URL: https://codereview.chromium.org/1469113003 .
parent 614140e2
Showing with 136 additions and 55 deletions
crosstest/test_arith.cpp
...@@ -49,3 +49,37 @@ SINTOP_TABLE ...@@ -49,3 +49,37 @@ SINTOP_TABLE
v4f32 test##inst(v4f32 a, v4f32 b) { return func(a op b); } v4f32 test##inst(v4f32 a, v4f32 b) { return func(a op b); }
FPOP_TABLE FPOP_TABLE
#undef X #undef X
#define X(mult_by) \
bool testMultiplyBy##mult_by(bool a, bool /*unused*/) { \
return a * (mult_by); \
} \
bool testMultiplyByNeg##mult_by(bool a, bool /*unused*/) { \
return a * (-(mult_by)); \
} \
uint8_t testMultiplyBy##mult_by(uint8_t a, uint8_t /*unused*/) { \
return a * (mult_by); \
} \
uint8_t testMultiplyByNeg##mult_by(uint8_t a, uint8_t /*unused*/) { \
return a * (-(mult_by)); \
} \
uint16_t testMultiplyBy##mult_by(uint16_t a, uint16_t /*unused*/) { \
return a * (mult_by); \
} \
uint16_t testMultiplyByNeg##mult_by(uint16_t a, uint16_t /*unused*/) { \
return a * (-(mult_by)); \
} \
uint32_t testMultiplyBy##mult_by(uint32_t a, uint32_t /*unused*/) { \
return a * (mult_by); \
} \
uint32_t testMultiplyByNeg##mult_by(uint32_t a, uint32_t /*unused*/) { \
return a * (-(mult_by)); \
} \
uint64_t testMultiplyBy##mult_by(uint64_t a, uint64_t /*unused*/) { \
return a * (mult_by); \
} \
uint64_t testMultiplyByNeg##mult_by(uint64_t a, uint64_t /*unused*/) { \
return a * (-(mult_by)); \
}
MULIMM_TABLE
#undef X
crosstest/test_arith.def
...@@ -17,35 +17,35 @@ ...@@ -17,35 +17,35 @@
#define XSTR(s) STR(s) #define XSTR(s) STR(s)
#define STR(s) #s #define STR(s) #s
#define UINTOP_TABLE \ #define UINTOP_TABLE \
/* inst, operator, div, shift */ \ /* inst, operator, div, shift */ \
X(Add, +, 0, 0) \ X(Add, +, 0, 0) \
X(Sub, -, 0, 0) \ X(Sub, -, 0, 0) \
X(Mul, *, 0, 0) \ X(Mul, *, 0, 0) \
X(Udiv, /, 1, 0) \ X(Udiv, /, 1, 0) \
X(Urem, %, 1, 0) \ X(Urem, %, 1, 0) \
X(Shl, <<, 0, 1) \ X(Shl, <<, 0, 1) \
X(Lshr, >>, 0, 1) \ X(Lshr, >>, 0, 1) \
X(And, &, 0, 0) \ X(And, &, 0, 0) \
X(Or, |, 0, 0) \ X(Or, |, 0, 0) \
X(Xor, ^, 0, 0) \ X(Xor, ^, 0, 0) \
//#define X(inst, op, isdiv, isshift) //#define X(inst, op, isdiv, isshift)
#define SINTOP_TABLE \ #define SINTOP_TABLE \
/* inst, operator, div, shift */ \ /* inst, operator, div, shift */ \
X(Sdiv, /, 1, 0) \ X(Sdiv, /, 1, 0) \
X(Srem, %, 1, 0) \ X(Srem, %, 1, 0) \
X(Ashr, >>, 0, 1) \ X(Ashr, >>, 0, 1) \
//#define X(inst, op, isdiv, isshift) //#define X(inst, op, isdiv, isshift)
#define COMMA , #define COMMA ,
#define FPOP_TABLE \ #define FPOP_TABLE \
/* inst, infix_op, func */ \ /* inst, infix_op, func */ \
X(Fadd, +, ) \ X(Fadd, +, ) \
X(Fsub, -, ) \ X(Fsub, -, ) \
X(Fmul, *, ) \ X(Fmul, *, ) \
X(Fdiv, /, ) \ X(Fdiv, /, ) \
X(Frem, COMMA, myFrem) \ X(Frem, COMMA, myFrem) \
//#define X(inst, op, func) //#define X(inst, op, func)
// Note: The above definition of COMMA, plus the "func" argument to // Note: The above definition of COMMA, plus the "func" argument to
...@@ -55,30 +55,51 @@ ...@@ -55,30 +55,51 @@
// instruction and "(a + b)" for the Fadd instruction. The two // instruction and "(a + b)" for the Fadd instruction. The two
// versions of myFrem() are defined in a separate bitcode file. // versions of myFrem() are defined in a separate bitcode file.
#define INT_VALUE_ARRAY \ #define INT_VALUE_ARRAY \
{ 0x0, 0x1, 0x7ffffffe, 0x7fffffff, \ { 0x0, 0x1, 0x7ffffffe, 0x7fffffff, \
0x80000000, 0x80000001, 0xfffffffe, 0xffffffff, \ 0x80000000, 0x80000001, 0xfffffffe, 0xffffffff, \
0x1e, 0x1f, 0x20, 0x21, 0x3e, 0x3f, 0x40, 0x41, \ 0x1e, 0x1f, 0x20, 0x21, 0x3e, 0x3f, 0x40, 0x41, \
0x7e, 0x7f, 0x80, 0x81, \ 0x7e, 0x7f, 0x80, 0x81, \
0xfe, 0xff, 0x100, 0x101, \ 0xfe, 0xff, 0x100, 0x101, \
0x7ffe, 0x7fff, 0x8000, 0x8001, \ 0x7ffe, 0x7fff, 0x8000, 0x8001, \
0xfffe, 0xffff, 0x10000, 0x10001 } 0xfffe, 0xffff, 0x10000, 0x10001 }
#define FP_VALUE_ARRAY(NegInf, PosInf, NegNan, NaN) \ #define FP_VALUE_ARRAY(NegInf, PosInf, NegNan, NaN) \
{ 0, 1, 1.4, \ { 0, 1, 1.4, \
1.5, 1.6, -1.4, \ 1.5, 1.6, -1.4, \
-1.5, -1.6, 0x7e, \ -1.5, -1.6, 0x7e, \
0x7f, 0x80, 0x81, \ 0x7f, 0x80, 0x81, \
0xfe, 0xff, 0x7ffe, \ 0xfe, 0xff, 0x7ffe, \
0x7fff, 0x8000, 0x8001, \ 0x7fff, 0x8000, 0x8001, \
0xfffe, 0xffff, 0x7ffffffe, \ 0xfffe, 0xffff, 0x7ffffffe, \
0x7fffffff, 0x80000000, 0x80000001, \ 0x7fffffff, 0x80000000, 0x80000001, \
0xfffffffe, 0xffffffff, 0x100000000ll, \ 0xfffffffe, 0xffffffff, 0x100000000ll, \
0x100000001ll, 0x7ffffffffffffffell, 0x7fffffffffffffffll, \ 0x100000001ll, 0x7ffffffffffffffell, 0x7fffffffffffffffll, \
0x8000000000000000ll, 0x8000000000000001ll, 0xfffffffffffffffell, \ 0x8000000000000000ll, 0x8000000000000001ll, 0xfffffffffffffffell, \
0xffffffffffffffffll, NegInf, PosInf, \ 0xffffffffffffffffll, NegInf, PosInf, \
Nan, NegNan, -0.0, \ Nan, NegNan, -0.0, \
10.0, FLT_MIN, FLT_MAX, \ 10.0, FLT_MIN, FLT_MAX, \
DBL_MIN, DBL_MAX } DBL_MIN, DBL_MAX }
#define MULIMM_TABLE \
/* mult_by */ \
X( 0) \
X( 1) \
X( 2) \
X( 3) \
X( 4) \
X( 5) \
X( 7) \
X( 8) \
X( 9) \
X( 10) \
X( 25) \
X( 100) \
X( 232) \
X(0x00FFF001) \
X(0x01000000) \
X(0x7FFFF07F) \
X(0x80000000) \
//#define X(mult_by)
#endif // TEST_ARITH_DEF #endif // TEST_ARITH_DEF
crosstest/test_arith.h
...@@ -60,3 +60,17 @@ double mySqrt(double a); ...@@ -60,3 +60,17 @@ double mySqrt(double a);
float myFabs(float a); float myFabs(float a);
double myFabs(double a); double myFabs(double a);
v4f32 myFabs(v4f32 a); v4f32 myFabs(v4f32 a);
#define X(mult_by) \
bool testMultiplyBy##mult_by(bool a, bool); \
bool testMultiplyByNeg##mult_by(bool a, bool); \
uint8_t testMultiplyBy##mult_by(uint8_t a, uint8_t); \
uint8_t testMultiplyByNeg##mult_by(uint8_t a, uint8_t); \
uint16_t testMultiplyBy##mult_by(uint16_t a, uint16_t); \
uint16_t testMultiplyByNeg##mult_by(uint16_t a, uint16_t); \
uint32_t testMultiplyBy##mult_by(uint32_t a, uint32_t); \
uint32_t testMultiplyByNeg##mult_by(uint32_t a, uint32_t); \
uint64_t testMultiplyBy##mult_by(uint64_t a, uint64_t); \
uint64_t testMultiplyByNeg##mult_by(uint64_t a, uint64_t);
MULIMM_TABLE
#undef X
crosstest/test_arith_main.cpp
...@@ -73,7 +73,15 @@ void testsInt(size_t &TotalTests, size_t &Passes, size_t &Failures) { ...@@ -73,7 +73,15 @@ void testsInt(size_t &TotalTests, size_t &Passes, size_t &Failures) {
, ,
SINTOP_TABLE SINTOP_TABLE
#undef X #undef X
}; #define X(mult_by) \
{ \
"Mult-By-" STR(mult_by), testMultiplyBy##mult_by, \
Subzero_::testMultiplyBy##mult_by, NULL, NULL, false \
} \
, {"Mult-By-Neg-" STR(mult_by), testMultiplyByNeg##mult_by, \
Subzero_::testMultiplyByNeg##mult_by, NULL, NULL, false},
MULIMM_TABLE};
#undef X
const static size_t NumFuncs = sizeof(Funcs) / sizeof(*Funcs); const static size_t NumFuncs = sizeof(Funcs) / sizeof(*Funcs);
if (sizeof(TypeUnsigned) <= sizeof(uint32_t)) { if (sizeof(TypeUnsigned) <= sizeof(uint32_t)) {
......
src/IceTargetLoweringARM32.h
...@@ -977,15 +977,16 @@ private: ...@@ -977,15 +977,16 @@ private:
// AllowTemporaryWithNoReg indicates if TargetARM32::makeReg() can be invoked // AllowTemporaryWithNoReg indicates if TargetARM32::makeReg() can be invoked
// without specifying a physical register. This is needed for creating unbound // without specifying a physical register. This is needed for creating unbound
// temporaries during Ice -> ARM lowering, but before register allocation. // temporaries during Ice -> ARM lowering, but before register allocation.
// This a safe-guard that, during the legalization post-passes no unbound // This a safe-guard that no unbound temporaries are created during the
// temporaries are created. // legalization post-passes.
bool AllowTemporaryWithNoReg = true; bool AllowTemporaryWithNoReg = true;
// ForbidTemporaryWithoutReg is a RAII class that manages // ForbidTemporaryWithoutReg is a RAII class that manages
// AllowTemporaryWithNoReg. // AllowTemporaryWithNoReg.
class ForbidTemporaryWithoutReg { class ForbidTemporaryWithoutReg {
ForbidTemporaryWithoutReg() = delete; ForbidTemporaryWithoutReg() = delete;
ForbidTemporaryWithoutReg(const ForbidTemporaryWithoutReg&) = delete; ForbidTemporaryWithoutReg(const ForbidTemporaryWithoutReg &) = delete;
ForbidTemporaryWithoutReg &operator=(const ForbidTemporaryWithoutReg&) = delete; ForbidTemporaryWithoutReg &
operator=(const ForbidTemporaryWithoutReg &) = delete;
public: public:
explicit ForbidTemporaryWithoutReg(TargetARM32 *Target) : Target(Target) { explicit ForbidTemporaryWithoutReg(TargetARM32 *Target) : Target(Target) {
......
tests_lit/llvm2ice_tests/arith.ll
...@@ -11,7 +11,7 @@ ...@@ -11,7 +11,7 @@
; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \ ; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \
; RUN: -i %s --args -O2 --skip-unimplemented \ ; RUN: -i %s --args -O2 --skip-unimplemented \
; RUN: | %if --need=target_ARM32 --need=allow_dump \ ; RUN: | %if --need=target_ARM32 --need=allow_dump \
; RUN: --command FileCheck --check-prefix ARM32 %s ; RUN: --command FileCheck --check-prefix ARM32 --check-prefix ARM-OPT2 %s
; RUN: %if --need=target_ARM32 --need=allow_dump \ ; RUN: %if --need=target_ARM32 --need=allow_dump \
; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \ ; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \
; RUN: -i %s --args -O2 --mattr=hwdiv-arm --skip-unimplemented \ ; RUN: -i %s --args -O2 --mattr=hwdiv-arm --skip-unimplemented \
...@@ -21,7 +21,7 @@ ...@@ -21,7 +21,7 @@
; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \ ; RUN: --command %p2i --filetype=asm --assemble --disassemble --target arm32 \
; RUN: -i %s --args -Om1 --skip-unimplemented \ ; RUN: -i %s --args -Om1 --skip-unimplemented \
; RUN: | %if --need=target_ARM32 --need=allow_dump \ ; RUN: | %if --need=target_ARM32 --need=allow_dump \
; RUN: --command FileCheck --check-prefix ARM32 %s ; RUN: --command FileCheck --check-prefix ARM32 --check-prefix ARM32-OPTM1 %s
; ;
; RUN: %if --need=target_MIPS32 --need=allow_dump \ ; RUN: %if --need=target_MIPS32 --need=allow_dump \
; RUN: --command %p2i --filetype=asm --assemble --disassemble --target mips32\ ; RUN: --command %p2i --filetype=asm --assemble --disassemble --target mips32\
...@@ -117,8 +117,11 @@ entry: ...@@ -117,8 +117,11 @@ entry:
; CHECK-LABEL: MulImm ; CHECK-LABEL: MulImm
; CHECK: imul e{{.*}},e{{.*}},0x63 ; CHECK: imul e{{.*}},e{{.*}},0x63
; ARM32-LABEL: MulImm ; ARM32-LABEL: MulImm
; ARM32: movw {{.*}}, #99 ; ARM32-OPTM1: movw {{.*}}, #99
; ARM32: mul r{{.*}}, r{{.*}}, r{{.*}} ; ARM32-OPTM1: mul r{{.*}}, r{{.*}}, r{{.*}}
; ARM32-OPT2: rsb [[T:r[0-9]+]], [[S:r[0-9]+]], [[S]], lsl #2
; ARM32-OPT2-DAG: add [[T]], [[T]], [[S]], lsl #7
; ARM32-OPT2-DAG: sub [[T]], [[T]], [[S]], lsl #5
; MIPS32-LABEL: MulImm ; MIPS32-LABEL: MulImm
; MIPS32: mul ; MIPS32: mul
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment