Skip to content

S
swiftshader
Commit 8d1072e7 by Matt Wala
Options

Lower vector floating point arithmetic operations.

This adds lowering code for fadd, fsub, fmul, fdiv, and frem. frem, having no native x86 counterpart, is implemented by making a helper call. BUG=none R=jvoung@chromium.org, stichnot@chromium.org Review URL: https://codereview.chromium.org/389653002
parent 78b4c0b8
Showing with 132 additions and 5 deletions
src/IceInstX8632.cpp
......@@ -399,9 +399,11 @@ void emitTwoAddress(const char *Opcode, const Inst *Inst, const Cfg *Func,
}
template <> const char *InstX8632Add::Opcode = "add";
template <> const char *InstX8632Addps::Opcode = "addps";
template <> const char *InstX8632Adc::Opcode = "adc";
template <> const char *InstX8632Addss::Opcode = "addss";
template <> const char *InstX8632Sub::Opcode = "sub";
template <> const char *InstX8632Subps::Opcode = "subps";
template <> const char *InstX8632Subss::Opcode = "subss";
template <> const char *InstX8632Sbb::Opcode = "sbb";
template <> const char *InstX8632And::Opcode = "and";
......@@ -409,8 +411,10 @@ template <> const char *InstX8632Or::Opcode = "or";
template <> const char *InstX8632Xor::Opcode = "xor";
template <> const char *InstX8632Pxor::Opcode = "pxor";
template <> const char *InstX8632Imul::Opcode = "imul";
template <> const char *InstX8632Mulps::Opcode = "mulps";
template <> const char *InstX8632Mulss::Opcode = "mulss";
template <> const char *InstX8632Div::Opcode = "div";
template <> const char *InstX8632Divps::Opcode = "divps";
template <> const char *InstX8632Idiv::Opcode = "idiv";
template <> const char *InstX8632Divss::Opcode = "divss";
template <> const char *InstX8632Shl::Opcode = "shl";
......
src/IceInstX8632.h
......@@ -136,6 +136,7 @@ public:
k__Start = Inst::Target,
Adc,
Add,
Addps,
Addss,
And,
Br,
......@@ -143,6 +144,7 @@ public:
Cdq,
Cvt,
Div,
Divps,
Divss,
Fld,
Fstp,
......@@ -158,6 +160,7 @@ public:
Movsx,
Movzx,
Mul,
Mulps,
Mulss,
Or,
Pop,
......@@ -174,6 +177,7 @@ public:
Store,
StoreQ,
Sub,
Subps,
Subss,
Test,
Ucomiss,
......@@ -397,9 +401,11 @@ private:
};
typedef InstX8632Binop<InstX8632::Add> InstX8632Add;
typedef InstX8632Binop<InstX8632::Addps> InstX8632Addps;
typedef InstX8632Binop<InstX8632::Adc> InstX8632Adc;
typedef InstX8632Binop<InstX8632::Addss> InstX8632Addss;
typedef InstX8632Binop<InstX8632::Sub> InstX8632Sub;
typedef InstX8632Binop<InstX8632::Subps> InstX8632Subps;
typedef InstX8632Binop<InstX8632::Subss> InstX8632Subss;
typedef InstX8632Binop<InstX8632::Sbb> InstX8632Sbb;
typedef InstX8632Binop<InstX8632::And> InstX8632And;
......@@ -407,7 +413,9 @@ typedef InstX8632Binop<InstX8632::Or> InstX8632Or;
typedef InstX8632Binop<InstX8632::Xor> InstX8632Xor;
typedef InstX8632Binop<InstX8632::Pxor> InstX8632Pxor;
typedef InstX8632Binop<InstX8632::Imul> InstX8632Imul;
typedef InstX8632Binop<InstX8632::Mulps> InstX8632Mulps;
typedef InstX8632Binop<InstX8632::Mulss> InstX8632Mulss;
typedef InstX8632Binop<InstX8632::Divps> InstX8632Divps;
typedef InstX8632Binop<InstX8632::Divss> InstX8632Divss;
typedef InstX8632Binop<InstX8632::Shl, true> InstX8632Shl;
typedef InstX8632Binop<InstX8632::Shr, true> InstX8632Shr;
......
src/IceTargetLoweringX8632.cpp
......@@ -1136,7 +1136,59 @@ void TargetX8632::lowerArithmetic(const InstArithmetic *Inst) {
llvm_unreachable("FP instruction with i64 type");
break;
}
} else { // Dest->getType() != IceType_i64
} else if (isVectorType(Dest->getType())) {
switch (Inst->getOp()) {
case InstArithmetic::_num:
llvm_unreachable("Unknown arithmetic operator");
break;
case InstArithmetic::Add:
case InstArithmetic::And:
case InstArithmetic::Or:
case InstArithmetic::Xor:
case InstArithmetic::Sub:
case InstArithmetic::Mul:
case InstArithmetic::Shl:
case InstArithmetic::Lshr:
case InstArithmetic::Ashr:
case InstArithmetic::Udiv:
case InstArithmetic::Sdiv:
case InstArithmetic::Urem:
case InstArithmetic::Srem:
// TODO(wala): Handle these.
Func->setError("Unhandled instruction");
break;
case InstArithmetic::Fadd: {
Variable *T = makeReg(Dest->getType());
_movp(T, Src0);
_addps(T, Src1);
_movp(Dest, T);
} break;
case InstArithmetic::Fsub: {
Variable *T = makeReg(Dest->getType());
_movp(T, Src0);
_subps(T, Src1);
_movp(Dest, T);
} break;
case InstArithmetic::Fmul: {
Variable *T = makeReg(Dest->getType());
_movp(T, Src0);
_mulps(T, Src1);
_movp(Dest, T);
} break;
case InstArithmetic::Fdiv: {
Variable *T = makeReg(Dest->getType());
_movp(T, Src0);
_divps(T, Src1);
_movp(Dest, T);
} break;
case InstArithmetic::Frem: {
const SizeT MaxSrcs = 1;
InstCall *Call = makeHelperCall("__frem_v4f32", Dest, MaxSrcs);
Call->addArg(Src0);
lowerCall(Call);
} break;
}
} else { // Dest->getType() is non-i64 scalar
Variable *T_edx = NULL;
Variable *T = NULL;
switch (Inst->getOp()) {
......@@ -1454,7 +1506,7 @@ void TargetX8632::lowerCall(const InstCall *Instr) {
Inst *FakeUse = InstFakeUse::create(Func, ReturnReg);
Context.insert(FakeUse);
}
if (!Dest)
return;
......
src/IceTargetLoweringX8632.h
......@@ -148,6 +148,9 @@ protected:
void _add(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Add::create(Func, Dest, Src0));
}
void _addps(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Addps::create(Func, Dest, Src0));
}
void _addss(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Addss::create(Func, Dest, Src0));
}
......@@ -180,6 +183,9 @@ protected:
void _div(Variable *Dest, Operand *Src0, Operand *Src1) {
Context.insert(InstX8632Div::create(Func, Dest, Src0, Src1));
}
void _divps(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Divps::create(Func, Dest, Src0));
}
void _divss(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Divss::create(Func, Dest, Src0));
}
......@@ -220,6 +226,9 @@ protected:
void _mul(Variable *Dest, Variable *Src0, Operand *Src1) {
Context.insert(InstX8632Mul::create(Func, Dest, Src0, Src1));
}
void _mulps(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Mulps::create(Func, Dest, Src0));
}
void _mulss(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Mulss::create(Func, Dest, Src0));
}
......@@ -232,6 +241,9 @@ protected:
void _push(Operand *Src0, bool SuppressStackAdjustment = false) {
Context.insert(InstX8632Push::create(Func, Src0, SuppressStackAdjustment));
}
void _pxor(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Pxor::create(Func, Dest, Src0));
}
void _ret(Variable *Src0 = NULL) {
Context.insert(InstX8632Ret::create(Func, Src0));
}
......@@ -265,6 +277,9 @@ protected:
void _sub(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Sub::create(Func, Dest, Src0));
}
void _subps(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Subps::create(Func, Dest, Src0));
}
void _subss(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Subss::create(Func, Dest, Src0));
}
......@@ -284,9 +299,6 @@ protected:
void _xor(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Xor::create(Func, Dest, Src0));
}
void _pxor(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Pxor::create(Func, Dest, Src0));
}
bool IsEbpBasedFrame;
size_t FrameSizeLocals;
......
tests_lit/llvm2ice_tests/vector-arith.ll 0 → 100644
; This test checks support for vector arithmetic.
; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s
; RUN: %llvm2ice -Om1 --verbose none %s | FileCheck %s
; RUN: %llvm2ice --verbose none %s | FileCheck --check-prefix=ERRORS %s
; RUN: %llvm2iceinsts %s | %szdiff %s | FileCheck --check-prefix=DUMP %s
; RUN: %llvm2iceinsts --pnacl %s | %szdiff %s \
; RUN: | FileCheck --check-prefix=DUMP %s
define <4 x float> @test_fadd(<4 x float> %arg0, <4 x float> %arg1) {
entry:
%res = fadd <4 x float> %arg0, %arg1
ret <4 x float> %res
; CHECK-LABEL: test_fadd:
; CHECK: addps
}
define <4 x float> @test_fsub(<4 x float> %arg0, <4 x float> %arg1) {
entry:
%res = fsub <4 x float> %arg0, %arg1
ret <4 x float> %res
; CHECK-LABEL: test_fsub:
; CHECK: subps
}
define <4 x float> @test_fmul(<4 x float> %arg0, <4 x float> %arg1) {
entry:
%res = fmul <4 x float> %arg0, %arg1
ret <4 x float> %res
; CHECK-LABEL: test_fmul:
; CHECK: mulps
}
define <4 x float> @test_fdiv(<4 x float> %arg0, <4 x float> %arg1) {
entry:
%res = fdiv <4 x float> %arg0, %arg1
ret <4 x float> %res
; CHECK-LABEL: test_fdiv:
; CHECK: divps
}
define <4 x float> @test_frem(<4 x float> %arg0, <4 x float> %arg1) {
entry:
%res = frem <4 x float> %arg0, %arg1
ret <4 x float> %res
; CHECK-LABEL: test_frem:
; CHECK: __frem_v4f32
}
; ERRORS-NOT: ICE translation error
; DUMP-NOT: SZ
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