gnh1201/wasm-micro-runtime
1
0
Fork 0
mirror of https://github.com/bytecodealliance/wasm-micro-runtime.git synced 2025-05-12 20:51:21 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Implement SIMD float ceil/floor/trunc/nearest opcodes (#472)

Browse Source 
Implement SIMD f32x4/f64x2 ceil/floor/trunc/nearest opcodes

Signed-off-by: Wenyong Huang <wenyong.huang@intel.com>
This commit is contained in:
Wenyong Huang 2020-12-16 01:33:25 -06:00 committed by GitHub
parent 16e6f41b3a
commit ad35c3c21b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 171 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
core/iwasm/compilation/aot_compiler.c
View File

@ -1741,6 +1741,55 @@ build_atomic_rmw:
break; break;
} }
case SIMD_f32x4_ceil:
{
if (!aot_compile_simd_f32x4_ceil(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f64x2_ceil:
{
if (!aot_compile_simd_f64x2_ceil(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f32x4_floor:
{
if (!aot_compile_simd_f32x4_floor(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f64x2_floor:
{
if (!aot_compile_simd_f64x2_floor(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f32x4_trunc:
{
if (!aot_compile_simd_f32x4_trunc(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f64x2_trunc:
{
if (!aot_compile_simd_f64x2_trunc(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f32x4_nearest:
{
if (!aot_compile_simd_f32x4_nearest(comp_ctx, func_ctx))
return false;
break;
}
case SIMD_f64x2_nearest:
{
if (!aot_compile_simd_f64x2_nearest(comp_ctx, func_ctx))
return false;
break;
}
default: default:
break; break;
} }

123
core/iwasm/compilation/simd/simd_floating_point.c
View File

@ -178,54 +178,7 @@ aot_compile_simd_f64x2_neg(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
} }
static bool static bool
simd_v128_float_abs(AOTCompContext *comp_ctx, simd_v128_float_intrinsic(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx,
LLVMTypeRef vector_type,
const char *intrinsic)
{
LLVMValueRef vector, result;
LLVMTypeRef param_types[1] = { vector_type };
if (!(vector = simd_pop_v128_and_bitcast(comp_ctx, func_ctx, vector_type,
"vec"))) {
goto fail;
}
if (!(result = aot_call_llvm_intrinsic(comp_ctx, intrinsic, vector_type,
param_types, 1, vector))) {
HANDLE_FAILURE("LLVMBuildCall");
goto fail;
}
if (!(result = LLVMBuildBitCast(comp_ctx->builder, result, V128_i64x2_TYPE,
"ret"))) {
HANDLE_FAILURE("LLVMBuildBitCast");
goto fail;
}
/* push result into the stack */
PUSH_V128(result);
return true;
fail:
return false;
}
bool
aot_compile_simd_f32x4_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_abs(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.fabs.v4f32");
}
bool
aot_compile_simd_f64x2_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_abs(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.fabs.v2f64");
}
static bool
simd_v128_float_sqrt(AOTCompContext *comp_ctx,
AOTFuncContext *func_ctx, AOTFuncContext *func_ctx,
LLVMTypeRef vector_type, LLVMTypeRef vector_type,
const char *intrinsic) const char *intrinsic)
@ -258,16 +211,86 @@ fail:
return false; return false;
} }
bool
aot_compile_simd_f32x4_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.fabs.v4f32");
}
bool
aot_compile_simd_f64x2_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.fabs.v2f64");
}
bool bool
aot_compile_simd_f32x4_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx) aot_compile_simd_f32x4_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{ {
return simd_v128_float_sqrt(comp_ctx, func_ctx, V128_f32x4_TYPE, return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.sqrt.v4f32"); "llvm.sqrt.v4f32");
} }
bool bool
aot_compile_simd_f64x2_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx) aot_compile_simd_f64x2_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{ {
return simd_v128_float_sqrt(comp_ctx, func_ctx, V128_f64x2_TYPE, return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.sqrt.v2f64"); "llvm.sqrt.v2f64");
} }
bool
aot_compile_simd_f32x4_ceil(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.ceil.v4f32");
}
bool
aot_compile_simd_f64x2_ceil(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.ceil.v2f64");
}
bool
aot_compile_simd_f32x4_floor(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.floor.v4f32");
}
bool
aot_compile_simd_f64x2_floor(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.floor.v2f64");
}
bool
aot_compile_simd_f32x4_trunc(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.trunc.v4f32");
}
bool
aot_compile_simd_f64x2_trunc(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.trunc.v2f64");
}
bool
aot_compile_simd_f32x4_nearest(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f32x4_TYPE,
"llvm.rint.v4f32");
}
bool
aot_compile_simd_f64x2_nearest(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
return simd_v128_float_intrinsic(comp_ctx, func_ctx, V128_f64x2_TYPE,
"llvm.rint.v2f64");
}

30
core/iwasm/compilation/simd/simd_floating_point.h
View File

@ -35,12 +35,34 @@ bool
aot_compile_simd_f64x2_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx); aot_compile_simd_f64x2_abs(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool bool
aot_compile_simd_f32x4_sqrt(AOTCompContext *comp_ctx, aot_compile_simd_f32x4_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
AOTFuncContext *func_ctx);
bool bool
aot_compile_simd_f64x2_sqrt(AOTCompContext *comp_ctx, aot_compile_simd_f64x2_sqrt(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
AOTFuncContext *func_ctx);
bool
aot_compile_simd_f32x4_ceil(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f64x2_ceil(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f32x4_floor(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f64x2_floor(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f32x4_trunc(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f64x2_trunc(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f32x4_nearest(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
bool
aot_compile_simd_f64x2_nearest(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);
#ifdef __cplusplus #ifdef __cplusplus
} /* end of extern "C" */ } /* end of extern "C" */

8
core/iwasm/interpreter/wasm_loader.c
View File

@ -7481,6 +7481,14 @@ fail_data_cnt_sec_require:
POP2_AND_PUSH(VALUE_TYPE_V128, VALUE_TYPE_V128); POP2_AND_PUSH(VALUE_TYPE_V128, VALUE_TYPE_V128);
break; break;
case SIMD_f32x4_ceil:
case SIMD_f32x4_floor:
case SIMD_f32x4_trunc:
case SIMD_f32x4_nearest:
case SIMD_f64x2_ceil:
case SIMD_f64x2_floor:
case SIMD_f64x2_trunc:
case SIMD_f64x2_nearest:
case SIMD_v128_not: case SIMD_v128_not:
case SIMD_i8x16_abs: case SIMD_i8x16_abs:
case SIMD_i8x16_neg: case SIMD_i8x16_neg:

10
core/iwasm/interpreter/wasm_opcode.h
View File

@ -472,6 +472,16 @@ typedef enum WASMSimdEXTOpcode {
SIMD_i64x2_sub = 0xd1, SIMD_i64x2_sub = 0xd1,
SIMD_i64x2_mul = 0xd5, SIMD_i64x2_mul = 0xd5,
/* float ceil/floor/trunc/nearest */
SIMD_f32x4_ceil = 0xd8,
SIMD_f32x4_floor = 0xd9,
SIMD_f32x4_trunc = 0xda,
SIMD_f32x4_nearest = 0xdb,
SIMD_f64x2_ceil = 0xdc,
SIMD_f64x2_floor = 0xdd,
SIMD_f64x2_trunc = 0xde,
SIMD_f64x2_nearest = 0xdf,
/* f32x4 operation */ /* f32x4 operation */
SIMD_f32x4_abs = 0xe0, SIMD_f32x4_abs = 0xe0,
SIMD_f32x4_neg = 0xe1, SIMD_f32x4_neg = 0xe1,