Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Patch implementations of vfbinop(min,max,pmin,pax) #2584

Merged
merged 1 commit into from
Sep 28, 2023
Merged

Patch implementations of vfbinop(min,max,pmin,pax) #2584

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
235 changes: 197 additions & 38 deletions core/iwasm/compilation/simd/simd_floating_point.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,10 +217,9 @@ aot_compile_simd_f64x2_nearest(AOTCompContext *comp_ctx,

static bool
simd_float_cmp(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
FloatArithmetic arith_op, LLVMTypeRef vector_type)
FloatArithmetic op, LLVMTypeRef vector_type)
{
LLVMValueRef lhs, rhs, result;
LLVMRealPredicate op = FLOAT_MIN == arith_op ? LLVMRealULT : LLVMRealUGT;
LLVMValueRef lhs, rhs, cmp, selected;

if (!(rhs =
simd_pop_v128_and_bitcast(comp_ctx, func_ctx, vector_type, "rhs"))
Expand All @@ -229,47 +228,122 @@ simd_float_cmp(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
return false;
}

if (!(result = LLVMBuildFCmp(comp_ctx->builder, op, lhs, rhs, "cmp"))) {
if (!(cmp = LLVMBuildFCmp(comp_ctx->builder,
op == FLOAT_MIN ? LLVMRealOLT : LLVMRealOGT, rhs,
lhs, "cmp"))) {
HANDLE_FAILURE("LLVMBuildFCmp");
return false;
}

if (!(result =
LLVMBuildSelect(comp_ctx->builder, result, lhs, rhs, "select"))) {
if (!(selected =
LLVMBuildSelect(comp_ctx->builder, cmp, rhs, lhs, "selected"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

return simd_bitcast_and_push_v128(comp_ctx, func_ctx, result, "result");
return simd_bitcast_and_push_v128(comp_ctx, func_ctx, selected, "result");
}

/*TODO: sugggest non-IA platforms check with "llvm.minimum.*" and
* "llvm.maximum.*" firstly */
bool
aot_compile_simd_f32x4_min_max(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
static bool
simd_float_min(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
LLVMTypeRef vector_type)
{
return simd_float_cmp(comp_ctx, func_ctx, run_min ? FLOAT_MIN : FLOAT_MAX,
V128_f32x4_TYPE);
}
LLVMValueRef lhs, rhs, lhs_nan, rhs_nan, olt_ret, ogt_ret, or_ret, ret1,
ret2, ret3, ret4;

bool
aot_compile_simd_f64x2_min_max(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
{
return simd_float_cmp(comp_ctx, func_ctx, run_min ? FLOAT_MIN : FLOAT_MAX,
V128_f64x2_TYPE);
if (!(rhs =
simd_pop_v128_and_bitcast(comp_ctx, func_ctx, vector_type, "rhs"))
|| !(lhs = simd_pop_v128_and_bitcast(comp_ctx, func_ctx, vector_type,
"lhs"))) {
return false;
}

if (!(lhs_nan = LLVMBuildFCmp(comp_ctx->builder, LLVMRealUNO, lhs, lhs,
"lhs_nan"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealUNO");
return false;
}

if (!(rhs_nan = LLVMBuildFCmp(comp_ctx->builder, LLVMRealUNO, rhs, rhs,
"rhs_nan"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealUNO");
return false;
}

if (!(olt_ret = LLVMBuildFCmp(comp_ctx->builder, LLVMRealOLT, lhs, rhs,
"olt_ret"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealOLT");
return false;
}

if (!(ogt_ret = LLVMBuildFCmp(comp_ctx->builder, LLVMRealOGT, lhs, rhs,
"ogt_ret"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealOGT");
return false;
}

/* lhs or rhs */
{
LLVMValueRef integer_l, integer_r, integer_or;

if (!(integer_l = LLVMBuildBitCast(comp_ctx->builder, lhs,
V128_i64x2_TYPE, "lhs_to_int"))) {
HANDLE_FAILURE("LLVMBuildBitCas");
return false;
}

if (!(integer_r = LLVMBuildBitCast(comp_ctx->builder, rhs,
V128_i64x2_TYPE, "rhs_to_int"))) {
HANDLE_FAILURE("LLVMBuildBitCas");
return false;
}

if (!(integer_or =
LLVMBuildOr(comp_ctx->builder, integer_l, integer_r, "or"))) {
HANDLE_FAILURE("LLVMBuildOr");
return false;
}

if (!(or_ret = LLVMBuildBitCast(comp_ctx->builder, integer_or,
vector_type, "holder"))) {
HANDLE_FAILURE("LLVMBuildBitCast");
return false;
}
}

if (!(ret1 = LLVMBuildSelect(comp_ctx->builder, olt_ret, lhs, or_ret,
"sel_olt"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret2 = LLVMBuildSelect(comp_ctx->builder, ogt_ret, rhs, ret1,
"sel_ogt"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret3 = LLVMBuildSelect(comp_ctx->builder, lhs_nan, lhs, ret2,
"sel_lhs_nan"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret4 = LLVMBuildSelect(comp_ctx->builder, rhs_nan, rhs, ret3,
"sel_rhs_nan"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

return simd_bitcast_and_push_v128(comp_ctx, func_ctx, ret4, "result");
}

static bool
simd_float_pmin_max(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
LLVMTypeRef vector_type, const char *intrinsic)
simd_float_max(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
LLVMTypeRef vector_type)
{
LLVMValueRef lhs, rhs, result;
LLVMTypeRef param_types[2];

param_types[0] = vector_type;
param_types[1] = vector_type;
LLVMValueRef lhs, rhs, lhs_nan, rhs_nan, olt_ret, ogt_ret, and_ret, ret1,
ret2, ret3, ret4;

if (!(rhs =
simd_pop_v128_and_bitcast(comp_ctx, func_ctx, vector_type, "rhs"))
Expand All @@ -278,31 +352,116 @@ simd_float_pmin_max(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
return false;
}

if (!(result =
aot_call_llvm_intrinsic(comp_ctx, func_ctx, intrinsic,
vector_type, param_types, 2, lhs, rhs))) {
if (!(lhs_nan = LLVMBuildFCmp(comp_ctx->builder, LLVMRealUNO, lhs, lhs,
"lhs_nan"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealUNO");
return false;
}

return simd_bitcast_and_push_v128(comp_ctx, func_ctx, result, "result");
if (!(rhs_nan = LLVMBuildFCmp(comp_ctx->builder, LLVMRealUNO, rhs, rhs,
"rhs_nan"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealUNO");
return false;
}

if (!(olt_ret = LLVMBuildFCmp(comp_ctx->builder, LLVMRealOLT, lhs, rhs,
"olt_ret"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealOLT");
return false;
}

if (!(ogt_ret = LLVMBuildFCmp(comp_ctx->builder, LLVMRealOGT, lhs, rhs,
"ogt_ret"))) {
HANDLE_FAILURE("LLVMBuildFCmp + LLVMRealOGT");
return false;
}

/* lhs and rhs */
{
LLVMValueRef integer_l, integer_r, integer_and;

if (!(integer_l = LLVMBuildBitCast(comp_ctx->builder, lhs,
V128_i64x2_TYPE, "lhs_to_int"))) {
HANDLE_FAILURE("LLVMBuildBitCas");
return false;
}

if (!(integer_r = LLVMBuildBitCast(comp_ctx->builder, rhs,
V128_i64x2_TYPE, "rhs_to_int"))) {
HANDLE_FAILURE("LLVMBuildBitCas");
return false;
}

if (!(integer_and = LLVMBuildAnd(comp_ctx->builder, integer_l,
integer_r, "and"))) {
HANDLE_FAILURE("LLVMBuildOr");
return false;
}

if (!(and_ret = LLVMBuildBitCast(comp_ctx->builder, integer_and,
vector_type, "holder"))) {
HANDLE_FAILURE("LLVMBuildBitCast");
return false;
}
}

if (!(ret1 = LLVMBuildSelect(comp_ctx->builder, ogt_ret, lhs, and_ret,
"sel_ogt"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret2 = LLVMBuildSelect(comp_ctx->builder, olt_ret, rhs, ret1,
"sel_olt"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret3 = LLVMBuildSelect(comp_ctx->builder, lhs_nan, lhs, ret2,
"sel_lhs_nan"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

if (!(ret4 = LLVMBuildSelect(comp_ctx->builder, rhs_nan, rhs, ret3,
"sel_rhs_nan"))) {
HANDLE_FAILURE("LLVMBuildSelect");
return false;
}

return simd_bitcast_and_push_v128(comp_ctx, func_ctx, ret4, "result");
}

bool
aot_compile_simd_f32x4_min_max(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
{
return run_min ? simd_float_min(comp_ctx, func_ctx, V128_f32x4_TYPE)
: simd_float_max(comp_ctx, func_ctx, V128_f32x4_TYPE);
}

bool
aot_compile_simd_f64x2_min_max(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
{
return run_min ? simd_float_min(comp_ctx, func_ctx, V128_f64x2_TYPE)
: simd_float_max(comp_ctx, func_ctx, V128_f64x2_TYPE);
}

bool
aot_compile_simd_f32x4_pmin_pmax(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
{
return simd_float_pmin_max(comp_ctx, func_ctx, V128_f32x4_TYPE,
run_min ? "llvm.minnum.v4f32"
: "llvm.maxnum.v4f32");
return simd_float_cmp(comp_ctx, func_ctx, run_min ? FLOAT_MIN : FLOAT_MAX,
V128_f32x4_TYPE);
}

bool
aot_compile_simd_f64x2_pmin_pmax(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, bool run_min)
{
return simd_float_pmin_max(comp_ctx, func_ctx, V128_f64x2_TYPE,
run_min ? "llvm.minnum.v2f64"
: "llvm.maxnum.v2f64");
return simd_float_cmp(comp_ctx, func_ctx, run_min ? FLOAT_MIN : FLOAT_MAX,
V128_f64x2_TYPE);
}

bool
Expand Down