mirror of
https://github.com/encounter/dawn-cmake.git
synced 2025-12-15 16:16:08 +00:00
Tint/E2E: Add f16 uniform/storage buffer E2E tests
This CL add Tint E2E tests for f16 types in uniform and storage buffers. Bug: tint:1473, tint:1502 Change-Id: I325524d2df326240cc1b080a90abf5bd076b3da1 Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/107543 Reviewed-by: Ben Clayton <bclayton@google.com> Kokoro: Kokoro <noreply+kokoro@google.com> Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
This commit is contained in:
Zhaoming Jiang
committed by
Dawn LUCI CQ
Dawn LUCI CQ
parent
205e16de63
commit
776b221ae2
@@ -0,0 +1,17 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> a : array<mat4x3<f16>, 4>;
|
||||
|
||||
var<private> counter = 0;
|
||||
fn i() -> i32 { counter++; return counter; }
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let p_a = &a;
|
||||
let p_a_i = &((*p_a)[i()]);
|
||||
let p_a_i_i = &((*p_a_i)[i()]);
|
||||
|
||||
let l_a : array<mat4x3<f16>, 4> = *p_a;
|
||||
let l_a_i : mat4x3<f16> = *p_a_i;
|
||||
let l_a_i_i : vec3<f16> = *p_a_i_i;
|
||||
}
|
||||
@@ -0,0 +1,59 @@
|
||||
cbuffer cbuffer_a : register(b0, space0) {
|
||||
uint4 a[8];
|
||||
};
|
||||
static int counter = 0;
|
||||
|
||||
int i() {
|
||||
counter = (counter + 1);
|
||||
return counter;
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_1(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const int p_a_i_save = i();
|
||||
const int p_a_i_i_save = i();
|
||||
const matrix<float16_t, 4, 3> l_a[4] = tint_symbol(a, 0u);
|
||||
const matrix<float16_t, 4, 3> l_a_i = tint_symbol_1(a, (32u * uint(p_a_i_save)));
|
||||
const uint scalar_offset_4 = (((32u * uint(p_a_i_save)) + (8u * uint(p_a_i_i_save)))) / 4;
|
||||
uint4 ubo_load_9 = a[scalar_offset_4 / 4];
|
||||
uint2 ubo_load_8 = ((scalar_offset_4 & 2) ? ubo_load_9.zw : ubo_load_9.xy);
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const vector<float16_t, 3> l_a_i_i = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,64 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_a : register(b0, space0) {
|
||||
uint4 a[8];
|
||||
};
|
||||
static int counter = 0;
|
||||
|
||||
int i() {
|
||||
counter = (counter + 1);
|
||||
return counter;
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_1(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const int p_a_i_save = i();
|
||||
const int p_a_i_i_save = i();
|
||||
const matrix<float16_t, 4, 3> l_a[4] = tint_symbol(a, 0u);
|
||||
const matrix<float16_t, 4, 3> l_a_i = tint_symbol_1(a, (32u * uint(p_a_i_save)));
|
||||
const uint scalar_offset_4 = (((32u * uint(p_a_i_save)) + (8u * uint(p_a_i_i_save)))) / 4;
|
||||
uint4 ubo_load_9 = a[scalar_offset_4 / 4];
|
||||
uint2 ubo_load_8 = ((scalar_offset_4 & 2) ? ubo_load_9.zw : ubo_load_9.xy);
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const vector<float16_t, 3> l_a_i_i = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]);
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000180EC89DCE0(11,8-16): error X3000: syntax error: unexpected token 'float16_t'
|
||||
|
||||
@@ -0,0 +1,75 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform a_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} a;
|
||||
|
||||
int counter = 0;
|
||||
int i() {
|
||||
counter = (counter + 1);
|
||||
return counter;
|
||||
}
|
||||
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
f16vec3 load_a_inner_p0_p1(uint p0, uint p1) {
|
||||
switch(p1) {
|
||||
case 0u: {
|
||||
return a.inner[p0].col0;
|
||||
break;
|
||||
}
|
||||
case 1u: {
|
||||
return a.inner[p0].col1;
|
||||
break;
|
||||
}
|
||||
case 2u: {
|
||||
return a.inner[p0].col2;
|
||||
break;
|
||||
}
|
||||
case 3u: {
|
||||
return a.inner[p0].col3;
|
||||
break;
|
||||
}
|
||||
default: {
|
||||
return f16vec3(0.0hf);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void f() {
|
||||
f16mat4x3 p_a[4] = conv_arr4_mat4x3_f16(a.inner);
|
||||
int tint_symbol = i();
|
||||
f16mat4x3 p_a_i = conv_mat4x3_f16(a.inner[tint_symbol]);
|
||||
int tint_symbol_1 = i();
|
||||
f16vec3 p_a_i_i = load_a_inner_p0_p1(uint(tint_symbol), uint(tint_symbol_1));
|
||||
f16mat4x3 l_a[4] = conv_arr4_mat4x3_f16(a.inner);
|
||||
f16mat4x3 l_a_i = conv_mat4x3_f16(a.inner[tint_symbol]);
|
||||
f16vec3 l_a_i_i = load_a_inner_p0_p1(uint(tint_symbol), uint(tint_symbol_1));
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,33 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
int i() {
|
||||
thread int tint_symbol_2 = 0;
|
||||
tint_symbol_2 = as_type<int>((as_type<uint>(tint_symbol_2) + as_type<uint>(1)));
|
||||
return tint_symbol_2;
|
||||
}
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol_3 [[buffer(0)]]) {
|
||||
int const tint_symbol = i();
|
||||
int const p_a_i_save = tint_symbol;
|
||||
int const tint_symbol_1 = i();
|
||||
int const p_a_i_i_save = tint_symbol_1;
|
||||
tint_array<half4x3, 4> const l_a = *(tint_symbol_3);
|
||||
half4x3 const l_a_i = (*(tint_symbol_3))[p_a_i_save];
|
||||
half3 const l_a_i_i = (*(tint_symbol_3))[p_a_i_save][p_a_i_i_save];
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,181 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 108
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %a_block_std140 "a_block_std140"
|
||||
OpMemberName %a_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %a "a"
|
||||
OpName %counter "counter"
|
||||
OpName %i "i"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i_0 "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %load_a_inner_p0_p1 "load_a_inner_p0_p1"
|
||||
OpName %p0 "p0"
|
||||
OpName %p1 "p1"
|
||||
OpName %f "f"
|
||||
OpDecorate %a_block_std140 Block
|
||||
OpMemberDecorate %a_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %a NonWritable
|
||||
OpDecorate %a DescriptorSet 0
|
||||
OpDecorate %a Binding 0
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%a_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_a_block_std140 = OpTypePointer Uniform %a_block_std140
|
||||
%a = OpVariable %_ptr_Uniform_a_block_std140 Uniform
|
||||
%int = OpTypeInt 32 1
|
||||
%11 = OpConstantNull %int
|
||||
%_ptr_Private_int = OpTypePointer Private %int
|
||||
%counter = OpVariable %_ptr_Private_int Private %11
|
||||
%14 = OpTypeFunction %int
|
||||
%int_1 = OpConstant %int 1
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%21 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%31 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%38 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%41 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%54 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%67 = OpTypeFunction %v3half %uint %uint
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%uint_3 = OpConstant %uint 3
|
||||
%90 = OpConstantNull %v3half
|
||||
%void = OpTypeVoid
|
||||
%91 = OpTypeFunction %void
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%i = OpFunction %int None %14
|
||||
%16 = OpLabel
|
||||
%17 = OpLoad %int %counter
|
||||
%19 = OpIAdd %int %17 %int_1
|
||||
OpStore %counter %19
|
||||
%20 = OpLoad %int %counter
|
||||
OpReturnValue %20
|
||||
OpFunctionEnd
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %21
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%25 = OpLabel
|
||||
%26 = OpCompositeExtract %v3half %val 0
|
||||
%27 = OpCompositeExtract %v3half %val 1
|
||||
%28 = OpCompositeExtract %v3half %val 2
|
||||
%29 = OpCompositeExtract %v3half %val 3
|
||||
%30 = OpCompositeConstruct %mat4v3half %26 %27 %28 %29
|
||||
OpReturnValue %30
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %31
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%35 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %38
|
||||
%i_0 = OpVariable %_ptr_Function_uint Function %41
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %54
|
||||
OpBranch %42
|
||||
%42 = OpLabel
|
||||
OpLoopMerge %43 %44 None
|
||||
OpBranch %45
|
||||
%45 = OpLabel
|
||||
%47 = OpLoad %uint %i_0
|
||||
%48 = OpULessThan %bool %47 %uint_4
|
||||
%46 = OpLogicalNot %bool %48
|
||||
OpSelectionMerge %50 None
|
||||
OpBranchConditional %46 %51 %50
|
||||
%51 = OpLabel
|
||||
OpBranch %43
|
||||
%50 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%55 = OpLoad %uint %i_0
|
||||
%57 = OpAccessChain %_ptr_Function_mat4v3half %arr %55
|
||||
%59 = OpLoad %uint %i_0
|
||||
%61 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %59
|
||||
%62 = OpLoad %mat4x3_f16 %61
|
||||
%58 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %62
|
||||
OpStore %57 %58
|
||||
OpBranch %44
|
||||
%44 = OpLabel
|
||||
%63 = OpLoad %uint %i_0
|
||||
%65 = OpIAdd %uint %63 %uint_1
|
||||
OpStore %i_0 %65
|
||||
OpBranch %42
|
||||
%43 = OpLabel
|
||||
%66 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %66
|
||||
OpFunctionEnd
|
||||
%load_a_inner_p0_p1 = OpFunction %v3half None %67
|
||||
%p0 = OpFunctionParameter %uint
|
||||
%p1 = OpFunctionParameter %uint
|
||||
%71 = OpLabel
|
||||
OpSelectionMerge %72 None
|
||||
OpSwitch %p1 %73 0 %74 1 %75 2 %76 3 %77
|
||||
%74 = OpLabel
|
||||
%80 = OpAccessChain %_ptr_Uniform_v3half %a %uint_0 %p0 %uint_0
|
||||
%81 = OpLoad %v3half %80
|
||||
OpReturnValue %81
|
||||
%75 = OpLabel
|
||||
%82 = OpAccessChain %_ptr_Uniform_v3half %a %uint_0 %p0 %uint_1
|
||||
%83 = OpLoad %v3half %82
|
||||
OpReturnValue %83
|
||||
%76 = OpLabel
|
||||
%85 = OpAccessChain %_ptr_Uniform_v3half %a %uint_0 %p0 %uint_2
|
||||
%86 = OpLoad %v3half %85
|
||||
OpReturnValue %86
|
||||
%77 = OpLabel
|
||||
%88 = OpAccessChain %_ptr_Uniform_v3half %a %uint_0 %p0 %uint_3
|
||||
%89 = OpLoad %v3half %88
|
||||
OpReturnValue %89
|
||||
%73 = OpLabel
|
||||
OpReturnValue %90
|
||||
%72 = OpLabel
|
||||
OpReturnValue %90
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %91
|
||||
%94 = OpLabel
|
||||
%95 = OpFunctionCall %int %i
|
||||
%96 = OpFunctionCall %int %i
|
||||
%99 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %a %uint_0
|
||||
%100 = OpLoad %_arr_mat4x3_f16_uint_4 %99
|
||||
%97 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %100
|
||||
%103 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %a %uint_0 %95
|
||||
%104 = OpLoad %mat4x3_f16 %103
|
||||
%101 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %104
|
||||
%106 = OpBitcast %uint %95
|
||||
%107 = OpBitcast %uint %96
|
||||
%105 = OpFunctionCall %v3half %load_a_inner_p0_p1 %106 %107
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,20 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> a : array<mat4x3<f16>, 4>;
|
||||
|
||||
var<private> counter = 0;
|
||||
|
||||
fn i() -> i32 {
|
||||
counter++;
|
||||
return counter;
|
||||
}
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let p_a = &(a);
|
||||
let p_a_i = &((*(p_a))[i()]);
|
||||
let p_a_i_i = &((*(p_a_i))[i()]);
|
||||
let l_a : array<mat4x3<f16>, 4> = *(p_a);
|
||||
let l_a_i : mat4x3<f16> = *(p_a_i);
|
||||
let l_a_i_i : vec3<f16> = *(p_a_i_i);
|
||||
}
|
||||
@@ -0,0 +1,14 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> a : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let p_a = &a;
|
||||
let p_a_2 = &((*p_a)[2]);
|
||||
let p_a_2_1 = &((*p_a_2)[1]);
|
||||
|
||||
let l_a : array<mat4x3<f16>, 4> = *p_a;
|
||||
let l_a_i : mat4x3<f16> = *p_a_2;
|
||||
let l_a_i_i : vec3<f16> = *p_a_2_1;
|
||||
}
|
||||
@@ -0,0 +1,49 @@
|
||||
cbuffer cbuffer_a : register(b0, space0) {
|
||||
uint4 a[8];
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const matrix<float16_t, 4, 3> l_a[4] = tint_symbol(a, 0u);
|
||||
const matrix<float16_t, 4, 3> l_a_i = tint_symbol_1(a, 64u);
|
||||
uint2 ubo_load_8 = a[4].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const vector<float16_t, 3> l_a_i_i = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,54 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_a : register(b0, space0) {
|
||||
uint4 a[8];
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const matrix<float16_t, 4, 3> l_a[4] = tint_symbol(a, 0u);
|
||||
const matrix<float16_t, 4, 3> l_a_i = tint_symbol_1(a, 64u);
|
||||
uint2 ubo_load_8 = a[4].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const vector<float16_t, 3> l_a_i_i = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]);
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x0000021B182DE800(5,8-16): error X3000: syntax error: unexpected token 'float16_t'
|
||||
|
||||
@@ -0,0 +1,42 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform a_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} a;
|
||||
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f() {
|
||||
f16mat4x3 p_a[4] = conv_arr4_mat4x3_f16(a.inner);
|
||||
f16mat4x3 p_a_2 = conv_mat4x3_f16(a.inner[2u]);
|
||||
f16vec3 p_a_2_1 = a.inner[2u].col1;
|
||||
f16mat4x3 l_a[4] = conv_arr4_mat4x3_f16(a.inner);
|
||||
f16mat4x3 l_a_i = conv_mat4x3_f16(a.inner[2u]);
|
||||
f16vec3 l_a_i_i = a.inner[2u].col1;
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,23 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol [[buffer(0)]]) {
|
||||
tint_array<half4x3, 4> const l_a = *(tint_symbol);
|
||||
half4x3 const l_a_i = (*(tint_symbol))[2];
|
||||
half3 const l_a_i_i = (*(tint_symbol))[2][1];
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,129 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 73
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %a_block_std140 "a_block_std140"
|
||||
OpMemberName %a_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %a "a"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %f "f"
|
||||
OpDecorate %a_block_std140 Block
|
||||
OpMemberDecorate %a_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %a NonWritable
|
||||
OpDecorate %a DescriptorSet 0
|
||||
OpDecorate %a Binding 0
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%a_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_a_block_std140 = OpTypePointer Uniform %a_block_std140
|
||||
%a = OpVariable %_ptr_Uniform_a_block_std140 Uniform
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%10 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%20 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%27 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%30 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%43 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%void = OpTypeVoid
|
||||
%56 = OpTypeFunction %void
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %10
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%14 = OpLabel
|
||||
%15 = OpCompositeExtract %v3half %val 0
|
||||
%16 = OpCompositeExtract %v3half %val 1
|
||||
%17 = OpCompositeExtract %v3half %val 2
|
||||
%18 = OpCompositeExtract %v3half %val 3
|
||||
%19 = OpCompositeConstruct %mat4v3half %15 %16 %17 %18
|
||||
OpReturnValue %19
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %20
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%24 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %27
|
||||
%i = OpVariable %_ptr_Function_uint Function %30
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %43
|
||||
OpBranch %31
|
||||
%31 = OpLabel
|
||||
OpLoopMerge %32 %33 None
|
||||
OpBranch %34
|
||||
%34 = OpLabel
|
||||
%36 = OpLoad %uint %i
|
||||
%37 = OpULessThan %bool %36 %uint_4
|
||||
%35 = OpLogicalNot %bool %37
|
||||
OpSelectionMerge %39 None
|
||||
OpBranchConditional %35 %40 %39
|
||||
%40 = OpLabel
|
||||
OpBranch %32
|
||||
%39 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%44 = OpLoad %uint %i
|
||||
%46 = OpAccessChain %_ptr_Function_mat4v3half %arr %44
|
||||
%48 = OpLoad %uint %i
|
||||
%50 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %48
|
||||
%51 = OpLoad %mat4x3_f16 %50
|
||||
%47 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %51
|
||||
OpStore %46 %47
|
||||
OpBranch %33
|
||||
%33 = OpLabel
|
||||
%52 = OpLoad %uint %i
|
||||
%54 = OpIAdd %uint %52 %uint_1
|
||||
OpStore %i %54
|
||||
OpBranch %31
|
||||
%32 = OpLabel
|
||||
%55 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %55
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %56
|
||||
%59 = OpLabel
|
||||
%63 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %a %uint_0
|
||||
%64 = OpLoad %_arr_mat4x3_f16_uint_4 %63
|
||||
%60 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %64
|
||||
%68 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %a %uint_0 %uint_2
|
||||
%69 = OpLoad %mat4x3_f16 %68
|
||||
%65 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %69
|
||||
%71 = OpAccessChain %_ptr_Uniform_v3half %a %uint_0 %uint_2 %uint_1
|
||||
%72 = OpLoad %v3half %71
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,13 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> a : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let p_a = &(a);
|
||||
let p_a_2 = &((*(p_a))[2]);
|
||||
let p_a_2_1 = &((*(p_a_2))[1]);
|
||||
let l_a : array<mat4x3<f16>, 4> = *(p_a);
|
||||
let l_a_i : mat4x3<f16> = *(p_a_2);
|
||||
let l_a_i_i : vec3<f16> = *(p_a_2_1);
|
||||
}
|
||||
@@ -0,0 +1,10 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let t = transpose(u[2]);
|
||||
let l = length(u[0][1].zxy);
|
||||
let a = abs(u[0][1].zxy.x);
|
||||
}
|
||||
@@ -0,0 +1,41 @@
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const matrix<float16_t, 3, 4> t = transpose(tint_symbol(u, 64u));
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const float16_t l = length(vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
uint2 ubo_load_9 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_9_xz = vector<float16_t, 2>(f16tof32(ubo_load_9 & 0xFFFF));
|
||||
float16_t ubo_load_9_y = f16tof32(ubo_load_9[0] >> 16);
|
||||
const float16_t a = abs(vector<float16_t, 3>(ubo_load_9_xz[0], ubo_load_9_y, ubo_load_9_xz[1]).zxy.x);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,46 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
const matrix<float16_t, 3, 4> t = transpose(tint_symbol(u, 64u));
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
const float16_t l = length(vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
uint2 ubo_load_9 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_9_xz = vector<float16_t, 2>(f16tof32(ubo_load_9 & 0xFFFF));
|
||||
float16_t ubo_load_9_y = f16tof32(ubo_load_9[0] >> 16);
|
||||
const float16_t a = abs(vector<float16_t, 3>(ubo_load_9_xz[0], ubo_load_9_y, ubo_load_9_xz[1]).zxy.x);
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x000002C9B1C5A400(5,8-16): error X3000: syntax error: unexpected token 'float16_t'
|
||||
|
||||
@@ -0,0 +1,29 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform u_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} u;
|
||||
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
void f() {
|
||||
f16mat3x4 t = transpose(conv_mat4x3_f16(u.inner[2u]));
|
||||
float16_t l = length(u.inner[0u].col1.zxy);
|
||||
float16_t a = abs(u.inner[0u].col1.zxy[0u]);
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,23 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol [[buffer(0)]]) {
|
||||
half3x4 const t = transpose((*(tint_symbol))[2]);
|
||||
half const l = length(half3((*(tint_symbol))[0][1]).zxy);
|
||||
half const a = fabs(half3((*(tint_symbol))[0][1]).zxy[0]);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,83 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 46
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
%34 = OpExtInstImport "GLSL.std.450"
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %u_block_std140 "u_block_std140"
|
||||
OpMemberName %u_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %u "u"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %f "f"
|
||||
OpDecorate %u_block_std140 Block
|
||||
OpMemberDecorate %u_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %u NonWritable
|
||||
OpDecorate %u DescriptorSet 0
|
||||
OpDecorate %u Binding 0
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%u_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_u_block_std140 = OpTypePointer Uniform %u_block_std140
|
||||
%u = OpVariable %_ptr_Uniform_u_block_std140 Uniform
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%10 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%void = OpTypeVoid
|
||||
%20 = OpTypeFunction %void
|
||||
%v4half = OpTypeVector %half 4
|
||||
%mat3v4half = OpTypeMatrix %v4half 3
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%35 = OpConstantNull %uint
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %10
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%14 = OpLabel
|
||||
%15 = OpCompositeExtract %v3half %val 0
|
||||
%16 = OpCompositeExtract %v3half %val 1
|
||||
%17 = OpCompositeExtract %v3half %val 2
|
||||
%18 = OpCompositeExtract %v3half %val 3
|
||||
%19 = OpCompositeConstruct %mat4v3half %15 %16 %17 %18
|
||||
OpReturnValue %19
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %20
|
||||
%23 = OpLabel
|
||||
%31 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %u %uint_0 %uint_2
|
||||
%32 = OpLoad %mat4x3_f16 %31
|
||||
%27 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %32
|
||||
%24 = OpTranspose %mat3v4half %27
|
||||
%38 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %35 %uint_1
|
||||
%39 = OpLoad %v3half %38
|
||||
%40 = OpVectorShuffle %v3half %39 %39 2 0 1
|
||||
%33 = OpExtInst %half %34 Length %40
|
||||
%42 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %35 %uint_1
|
||||
%43 = OpLoad %v3half %42
|
||||
%44 = OpVectorShuffle %v3half %43 %43 2 0 1
|
||||
%45 = OpCompositeExtract %half %44 0
|
||||
%41 = OpExtInst %half %34 FAbs %45
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,10 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
let t = transpose(u[2]);
|
||||
let l = length(u[0][1].zxy);
|
||||
let a = abs(u[0][1].zxy.x);
|
||||
}
|
||||
16
test/tint/buffer/uniform/std140/array/mat4x3_f16/to_fn.wgsl
Normal file
16
test/tint/buffer/uniform/std140/array/mat4x3_f16/to_fn.wgsl
Normal file
@@ -0,0 +1,16 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
fn a(a : array<mat4x3<f16>, 4>) {}
|
||||
fn b(m : mat4x3<f16>) {}
|
||||
fn c(v : vec3<f16>) {}
|
||||
fn d(f : f16) {}
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
a(u);
|
||||
b(u[1]);
|
||||
c(u[1][0].zxy);
|
||||
d(u[1][0].zxy.x);
|
||||
}
|
||||
@@ -0,0 +1,65 @@
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
|
||||
void a(matrix<float16_t, 4, 3> a_1[4]) {
|
||||
}
|
||||
|
||||
void b(matrix<float16_t, 4, 3> m) {
|
||||
}
|
||||
|
||||
void c(vector<float16_t, 3> v) {
|
||||
}
|
||||
|
||||
void d(float16_t f_1) {
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
a(tint_symbol(u, 0u));
|
||||
b(tint_symbol_1(u, 32u));
|
||||
uint2 ubo_load_8 = u[2].xy;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
c(vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
uint2 ubo_load_9 = u[2].xy;
|
||||
vector<float16_t, 2> ubo_load_9_xz = vector<float16_t, 2>(f16tof32(ubo_load_9 & 0xFFFF));
|
||||
float16_t ubo_load_9_y = f16tof32(ubo_load_9[0] >> 16);
|
||||
d(vector<float16_t, 3>(ubo_load_9_xz[0], ubo_load_9_y, ubo_load_9_xz[1]).zxy.x);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,73 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
|
||||
void a(matrix<float16_t, 4, 3> a_1[4]) {
|
||||
}
|
||||
|
||||
void b(matrix<float16_t, 4, 3> m) {
|
||||
}
|
||||
|
||||
void c(vector<float16_t, 3> v) {
|
||||
}
|
||||
|
||||
void d(float16_t f_1) {
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
a(tint_symbol(u, 0u));
|
||||
b(tint_symbol_1(u, 32u));
|
||||
uint2 ubo_load_8 = u[2].xy;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
c(vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
uint2 ubo_load_9 = u[2].xy;
|
||||
vector<float16_t, 2> ubo_load_9_xz = vector<float16_t, 2>(f16tof32(ubo_load_9 & 0xFFFF));
|
||||
float16_t ubo_load_9_y = f16tof32(ubo_load_9[0] >> 16);
|
||||
d(vector<float16_t, 3>(ubo_load_9_xz[0], ubo_load_9_y, ubo_load_9_xz[1]).zxy.x);
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000296784C0E70(5,15-23): error X3000: syntax error: unexpected token 'float16_t'
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000296784C0E70(8,15-23): error X3000: syntax error: unexpected token 'float16_t'
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000296784C0E70(11,15-23): error X3000: syntax error: unexpected token 'float16_t'
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000296784C0E70(14,8-16): error X3000: unrecognized identifier 'float16_t'
|
||||
|
||||
@@ -0,0 +1,52 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform u_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} u;
|
||||
|
||||
void a(f16mat4x3 a_1[4]) {
|
||||
}
|
||||
|
||||
void b(f16mat4x3 m) {
|
||||
}
|
||||
|
||||
void c(f16vec3 v) {
|
||||
}
|
||||
|
||||
void d(float16_t f_1) {
|
||||
}
|
||||
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f() {
|
||||
a(conv_arr4_mat4x3_f16(u.inner));
|
||||
b(conv_mat4x3_f16(u.inner[1u]));
|
||||
c(u.inner[1u].col0.zxy);
|
||||
d(u.inner[1u].col0.zxy[0u]);
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,36 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
void a(tint_array<half4x3, 4> a_1) {
|
||||
}
|
||||
|
||||
void b(half4x3 m) {
|
||||
}
|
||||
|
||||
void c(half3 v) {
|
||||
}
|
||||
|
||||
void d(half f_1) {
|
||||
}
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol [[buffer(0)]]) {
|
||||
a(*(tint_symbol));
|
||||
b((*(tint_symbol))[1]);
|
||||
c(half3((*(tint_symbol))[1][0]).zxy);
|
||||
d(half3((*(tint_symbol))[1][0]).zxy[0]);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,169 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 97
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %u_block_std140 "u_block_std140"
|
||||
OpMemberName %u_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %u "u"
|
||||
OpName %a "a"
|
||||
OpName %a_1 "a_1"
|
||||
OpName %b "b"
|
||||
OpName %m "m"
|
||||
OpName %c "c"
|
||||
OpName %v "v"
|
||||
OpName %d "d"
|
||||
OpName %f_1 "f_1"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %f "f"
|
||||
OpDecorate %u_block_std140 Block
|
||||
OpMemberDecorate %u_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %u NonWritable
|
||||
OpDecorate %u DescriptorSet 0
|
||||
OpDecorate %u Binding 0
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%u_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_u_block_std140 = OpTypePointer Uniform %u_block_std140
|
||||
%u = OpVariable %_ptr_Uniform_u_block_std140 Uniform
|
||||
%void = OpTypeVoid
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%10 = OpTypeFunction %void %_arr_mat4v3half_uint_4
|
||||
%17 = OpTypeFunction %void %mat4v3half
|
||||
%21 = OpTypeFunction %void %v3half
|
||||
%25 = OpTypeFunction %void %half
|
||||
%29 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%38 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%44 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%47 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%60 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%73 = OpTypeFunction %void
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%a = OpFunction %void None %10
|
||||
%a_1 = OpFunctionParameter %_arr_mat4v3half_uint_4
|
||||
%16 = OpLabel
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
%b = OpFunction %void None %17
|
||||
%m = OpFunctionParameter %mat4v3half
|
||||
%20 = OpLabel
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
%c = OpFunction %void None %21
|
||||
%v = OpFunctionParameter %v3half
|
||||
%24 = OpLabel
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
%d = OpFunction %void None %25
|
||||
%f_1 = OpFunctionParameter %half
|
||||
%28 = OpLabel
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %29
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%32 = OpLabel
|
||||
%33 = OpCompositeExtract %v3half %val 0
|
||||
%34 = OpCompositeExtract %v3half %val 1
|
||||
%35 = OpCompositeExtract %v3half %val 2
|
||||
%36 = OpCompositeExtract %v3half %val 3
|
||||
%37 = OpCompositeConstruct %mat4v3half %33 %34 %35 %36
|
||||
OpReturnValue %37
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %38
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%41 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %44
|
||||
%i = OpVariable %_ptr_Function_uint Function %47
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %60
|
||||
OpBranch %48
|
||||
%48 = OpLabel
|
||||
OpLoopMerge %49 %50 None
|
||||
OpBranch %51
|
||||
%51 = OpLabel
|
||||
%53 = OpLoad %uint %i
|
||||
%54 = OpULessThan %bool %53 %uint_4
|
||||
%52 = OpLogicalNot %bool %54
|
||||
OpSelectionMerge %56 None
|
||||
OpBranchConditional %52 %57 %56
|
||||
%57 = OpLabel
|
||||
OpBranch %49
|
||||
%56 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%61 = OpLoad %uint %i
|
||||
%63 = OpAccessChain %_ptr_Function_mat4v3half %arr %61
|
||||
%65 = OpLoad %uint %i
|
||||
%67 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %65
|
||||
%68 = OpLoad %mat4x3_f16 %67
|
||||
%64 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %68
|
||||
OpStore %63 %64
|
||||
OpBranch %50
|
||||
%50 = OpLabel
|
||||
%69 = OpLoad %uint %i
|
||||
%71 = OpIAdd %uint %69 %uint_1
|
||||
OpStore %i %71
|
||||
OpBranch %48
|
||||
%49 = OpLabel
|
||||
%72 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %72
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %73
|
||||
%75 = OpLabel
|
||||
%80 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %u %uint_0
|
||||
%81 = OpLoad %_arr_mat4x3_f16_uint_4 %80
|
||||
%77 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %81
|
||||
%76 = OpFunctionCall %void %a %77
|
||||
%85 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %u %uint_0 %uint_1
|
||||
%86 = OpLoad %mat4x3_f16 %85
|
||||
%83 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %86
|
||||
%82 = OpFunctionCall %void %b %83
|
||||
%89 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %uint_1 %uint_0
|
||||
%90 = OpLoad %v3half %89
|
||||
%91 = OpVectorShuffle %v3half %90 %90 2 0 1
|
||||
%87 = OpFunctionCall %void %c %91
|
||||
%93 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %uint_1 %uint_0
|
||||
%94 = OpLoad %v3half %93
|
||||
%95 = OpVectorShuffle %v3half %94 %94 2 0 1
|
||||
%96 = OpCompositeExtract %half %95 0
|
||||
%92 = OpFunctionCall %void %d %96
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,23 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
fn a(a : array<mat4x3<f16>, 4>) {
|
||||
}
|
||||
|
||||
fn b(m : mat4x3<f16>) {
|
||||
}
|
||||
|
||||
fn c(v : vec3<f16>) {
|
||||
}
|
||||
|
||||
fn d(f : f16) {
|
||||
}
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
a(u);
|
||||
b(u[1]);
|
||||
c(u[1][0].zxy);
|
||||
d(u[1][0].zxy.x);
|
||||
}
|
||||
@@ -0,0 +1,12 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
var<private> p : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
p = u;
|
||||
p[1] = u[2];
|
||||
p[1][0] = u[0][1].zxy;
|
||||
p[1][0].x = u[0][1].x;
|
||||
}
|
||||
@@ -0,0 +1,51 @@
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
static matrix<float16_t, 4, 3> p[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
p = tint_symbol(u, 0u);
|
||||
p[1] = tint_symbol_1(u, 64u);
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
p[1][0] = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy;
|
||||
p[1][0].x = float16_t(f16tof32(((u[0].z) & 0xFFFF)));
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,56 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
static matrix<float16_t, 4, 3> p[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_1(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_ret[4];
|
||||
tint_symbol_ret tint_symbol(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = tint_symbol_1(buffer, (offset + (i * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
p = tint_symbol(u, 0u);
|
||||
p[1] = tint_symbol_1(u, 64u);
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
p[1][0] = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy;
|
||||
p[1][0].x = float16_t(f16tof32(((u[0].z) & 0xFFFF)));
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x00000222F970C590(4,15-23): error X3000: syntax error: unexpected token 'float16_t'
|
||||
|
||||
@@ -0,0 +1,41 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform u_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} u;
|
||||
|
||||
f16mat4x3 p[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f() {
|
||||
p = conv_arr4_mat4x3_f16(u.inner);
|
||||
p[1] = conv_mat4x3_f16(u.inner[2u]);
|
||||
p[1][0] = u.inner[0u].col1.zxy;
|
||||
p[1][0].x = u.inner[0u].col1[0u];
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,25 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol_1 [[buffer(0)]]) {
|
||||
thread tint_array<half4x3, 4> tint_symbol = {};
|
||||
tint_symbol = *(tint_symbol_1);
|
||||
tint_symbol[1] = (*(tint_symbol_1))[2];
|
||||
tint_symbol[1][0] = half3((*(tint_symbol_1))[0][1]).zxy;
|
||||
tint_symbol[1][0][0] = (*(tint_symbol_1))[0][1][0];
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,149 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 88
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %u_block_std140 "u_block_std140"
|
||||
OpMemberName %u_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %u "u"
|
||||
OpName %p "p"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %f "f"
|
||||
OpDecorate %u_block_std140 Block
|
||||
OpMemberDecorate %u_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %u NonWritable
|
||||
OpDecorate %u DescriptorSet 0
|
||||
OpDecorate %u Binding 0
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%u_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_u_block_std140 = OpTypePointer Uniform %u_block_std140
|
||||
%u = OpVariable %_ptr_Uniform_u_block_std140 Uniform
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%_ptr_Private__arr_mat4v3half_uint_4 = OpTypePointer Private %_arr_mat4v3half_uint_4
|
||||
%14 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%p = OpVariable %_ptr_Private__arr_mat4v3half_uint_4 Private %14
|
||||
%15 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%24 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%32 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%45 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%void = OpTypeVoid
|
||||
%58 = OpTypeFunction %void
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%int = OpTypeInt 32 1
|
||||
%int_1 = OpConstant %int 1
|
||||
%_ptr_Private_mat4v3half = OpTypePointer Private %mat4v3half
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%76 = OpConstantNull %int
|
||||
%_ptr_Private_v3half = OpTypePointer Private %v3half
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%_ptr_Private_half = OpTypePointer Private %half
|
||||
%_ptr_Uniform_half = OpTypePointer Uniform %half
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %15
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%18 = OpLabel
|
||||
%19 = OpCompositeExtract %v3half %val 0
|
||||
%20 = OpCompositeExtract %v3half %val 1
|
||||
%21 = OpCompositeExtract %v3half %val 2
|
||||
%22 = OpCompositeExtract %v3half %val 3
|
||||
%23 = OpCompositeConstruct %mat4v3half %19 %20 %21 %22
|
||||
OpReturnValue %23
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %24
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%27 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %14
|
||||
%i = OpVariable %_ptr_Function_uint Function %32
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %45
|
||||
OpBranch %33
|
||||
%33 = OpLabel
|
||||
OpLoopMerge %34 %35 None
|
||||
OpBranch %36
|
||||
%36 = OpLabel
|
||||
%38 = OpLoad %uint %i
|
||||
%39 = OpULessThan %bool %38 %uint_4
|
||||
%37 = OpLogicalNot %bool %39
|
||||
OpSelectionMerge %41 None
|
||||
OpBranchConditional %37 %42 %41
|
||||
%42 = OpLabel
|
||||
OpBranch %34
|
||||
%41 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%46 = OpLoad %uint %i
|
||||
%48 = OpAccessChain %_ptr_Function_mat4v3half %arr %46
|
||||
%50 = OpLoad %uint %i
|
||||
%52 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %50
|
||||
%53 = OpLoad %mat4x3_f16 %52
|
||||
%49 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %53
|
||||
OpStore %48 %49
|
||||
OpBranch %35
|
||||
%35 = OpLabel
|
||||
%54 = OpLoad %uint %i
|
||||
%56 = OpIAdd %uint %54 %uint_1
|
||||
OpStore %i %56
|
||||
OpBranch %33
|
||||
%34 = OpLabel
|
||||
%57 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %57
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %58
|
||||
%61 = OpLabel
|
||||
%65 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %u %uint_0
|
||||
%66 = OpLoad %_arr_mat4x3_f16_uint_4 %65
|
||||
%62 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %66
|
||||
OpStore %p %62
|
||||
%70 = OpAccessChain %_ptr_Private_mat4v3half %p %int_1
|
||||
%74 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %u %uint_0 %uint_2
|
||||
%75 = OpLoad %mat4x3_f16 %74
|
||||
%71 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %75
|
||||
OpStore %70 %71
|
||||
%78 = OpAccessChain %_ptr_Private_v3half %p %int_1 %76
|
||||
%80 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %32 %uint_1
|
||||
%81 = OpLoad %v3half %80
|
||||
%82 = OpVectorShuffle %v3half %81 %81 2 0 1
|
||||
OpStore %78 %82
|
||||
%84 = OpAccessChain %_ptr_Private_half %p %int_1 %76 %uint_0
|
||||
%86 = OpAccessChain %_ptr_Uniform_half %u %uint_0 %32 %uint_1 %32
|
||||
%87 = OpLoad %half %86
|
||||
OpStore %84 %87
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,13 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
var<private> p : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
p = u;
|
||||
p[1] = u[2];
|
||||
p[1][0] = u[0][1].zxy;
|
||||
p[1][0].x = u[0][1].x;
|
||||
}
|
||||
@@ -0,0 +1,12 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
@group(0) @binding(1) var<storage, read_write> s : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
s = u;
|
||||
s[1] = u[2];
|
||||
s[1][0] = u[0][1].zxy;
|
||||
s[1][0].x = u[0][1].x;
|
||||
}
|
||||
@@ -0,0 +1,67 @@
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
RWByteAddressBuffer s : register(u1, space0);
|
||||
|
||||
void tint_symbol_1(RWByteAddressBuffer buffer, uint offset, matrix<float16_t, 4, 3> value) {
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 0u), value[0u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 8u), value[1u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 16u), value[2u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 24u), value[3u]);
|
||||
}
|
||||
|
||||
void tint_symbol(RWByteAddressBuffer buffer, uint offset, matrix<float16_t, 4, 3> value[4]) {
|
||||
matrix<float16_t, 4, 3> array[4] = value;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
tint_symbol_1(buffer, (offset + (i * 32u)), array[i]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_4(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_3_ret[4];
|
||||
tint_symbol_3_ret tint_symbol_3(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_4(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
tint_symbol(s, 0u, tint_symbol_3(u, 0u));
|
||||
tint_symbol_1(s, 32u, tint_symbol_4(u, 64u));
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
s.Store<vector<float16_t, 3> >(32u, vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
s.Store<float16_t>(32u, float16_t(f16tof32(((u[0].z) & 0xFFFF))));
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,73 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
RWByteAddressBuffer s : register(u1, space0);
|
||||
|
||||
void tint_symbol_1(RWByteAddressBuffer buffer, uint offset, matrix<float16_t, 4, 3> value) {
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 0u), value[0u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 8u), value[1u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 16u), value[2u]);
|
||||
buffer.Store<vector<float16_t, 3> >((offset + 24u), value[3u]);
|
||||
}
|
||||
|
||||
void tint_symbol(RWByteAddressBuffer buffer, uint offset, matrix<float16_t, 4, 3> value[4]) {
|
||||
matrix<float16_t, 4, 3> array[4] = value;
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
tint_symbol_1(buffer, (offset + (i * 32u)), array[i]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_4(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_3_ret[4];
|
||||
tint_symbol_3_ret tint_symbol_3(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_4(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f() {
|
||||
tint_symbol(s, 0u, tint_symbol_3(u, 0u));
|
||||
tint_symbol_1(s, 32u, tint_symbol_4(u, 64u));
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
s.Store<vector<float16_t, 3> >(32u, vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy);
|
||||
s.Store<float16_t>(32u, float16_t(f16tof32(((u[0].z) & 0xFFFF))));
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x0000021F13DB33D0(6,68-76): error X3000: syntax error: unexpected token 'float16_t'
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x0000021F13DB33D0(7,3-14): error X3018: invalid subscript 'Store'
|
||||
|
||||
@@ -0,0 +1,44 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform u_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} u;
|
||||
|
||||
layout(binding = 1, std430) buffer u_block_ssbo {
|
||||
f16mat4x3 inner[4];
|
||||
} s;
|
||||
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f() {
|
||||
s.inner = conv_arr4_mat4x3_f16(u.inner);
|
||||
s.inner[1] = conv_mat4x3_f16(u.inner[2u]);
|
||||
s.inner[1][0] = u.inner[0u].col1.zxy;
|
||||
s.inner[1][0].x = u.inner[0u].col1[0u];
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f();
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,24 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
kernel void f(device tint_array<half4x3, 4>* tint_symbol [[buffer(1)]], const constant tint_array<half4x3, 4>* tint_symbol_1 [[buffer(0)]]) {
|
||||
*(tint_symbol) = *(tint_symbol_1);
|
||||
(*(tint_symbol))[1] = (*(tint_symbol_1))[2];
|
||||
(*(tint_symbol))[1][0] = half3((*(tint_symbol_1))[0][1]).zxy;
|
||||
(*(tint_symbol))[1][0][0] = (*(tint_symbol_1))[0][1][0];
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,160 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 91
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f"
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %u_block_std140 "u_block_std140"
|
||||
OpMemberName %u_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %u "u"
|
||||
OpName %u_block "u_block"
|
||||
OpMemberName %u_block 0 "inner"
|
||||
OpName %s "s"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %f "f"
|
||||
OpDecorate %u_block_std140 Block
|
||||
OpMemberDecorate %u_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %u NonWritable
|
||||
OpDecorate %u DescriptorSet 0
|
||||
OpDecorate %u Binding 0
|
||||
OpDecorate %u_block Block
|
||||
OpMemberDecorate %u_block 0 Offset 0
|
||||
OpMemberDecorate %u_block 0 ColMajor
|
||||
OpMemberDecorate %u_block 0 MatrixStride 8
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
OpDecorate %s DescriptorSet 0
|
||||
OpDecorate %s Binding 1
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint = OpTypeInt 32 0
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%u_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_u_block_std140 = OpTypePointer Uniform %u_block_std140
|
||||
%u = OpVariable %_ptr_Uniform_u_block_std140 Uniform
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%u_block = OpTypeStruct %_arr_mat4v3half_uint_4
|
||||
%_ptr_StorageBuffer_u_block = OpTypePointer StorageBuffer %u_block
|
||||
%s = OpVariable %_ptr_StorageBuffer_u_block StorageBuffer
|
||||
%15 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%24 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%30 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%33 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%46 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%void = OpTypeVoid
|
||||
%59 = OpTypeFunction %void
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_StorageBuffer__arr_mat4v3half_uint_4 = OpTypePointer StorageBuffer %_arr_mat4v3half_uint_4
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%int = OpTypeInt 32 1
|
||||
%int_1 = OpConstant %int 1
|
||||
%_ptr_StorageBuffer_mat4v3half = OpTypePointer StorageBuffer %mat4v3half
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%79 = OpConstantNull %int
|
||||
%_ptr_StorageBuffer_v3half = OpTypePointer StorageBuffer %v3half
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%_ptr_StorageBuffer_half = OpTypePointer StorageBuffer %half
|
||||
%_ptr_Uniform_half = OpTypePointer Uniform %half
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %15
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%18 = OpLabel
|
||||
%19 = OpCompositeExtract %v3half %val 0
|
||||
%20 = OpCompositeExtract %v3half %val 1
|
||||
%21 = OpCompositeExtract %v3half %val 2
|
||||
%22 = OpCompositeExtract %v3half %val 3
|
||||
%23 = OpCompositeConstruct %mat4v3half %19 %20 %21 %22
|
||||
OpReturnValue %23
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %24
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%27 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %30
|
||||
%i = OpVariable %_ptr_Function_uint Function %33
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %46
|
||||
OpBranch %34
|
||||
%34 = OpLabel
|
||||
OpLoopMerge %35 %36 None
|
||||
OpBranch %37
|
||||
%37 = OpLabel
|
||||
%39 = OpLoad %uint %i
|
||||
%40 = OpULessThan %bool %39 %uint_4
|
||||
%38 = OpLogicalNot %bool %40
|
||||
OpSelectionMerge %42 None
|
||||
OpBranchConditional %38 %43 %42
|
||||
%43 = OpLabel
|
||||
OpBranch %35
|
||||
%42 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%47 = OpLoad %uint %i
|
||||
%49 = OpAccessChain %_ptr_Function_mat4v3half %arr %47
|
||||
%51 = OpLoad %uint %i
|
||||
%53 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %51
|
||||
%54 = OpLoad %mat4x3_f16 %53
|
||||
%50 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %54
|
||||
OpStore %49 %50
|
||||
OpBranch %36
|
||||
%36 = OpLabel
|
||||
%55 = OpLoad %uint %i
|
||||
%57 = OpIAdd %uint %55 %uint_1
|
||||
OpStore %i %57
|
||||
OpBranch %34
|
||||
%35 = OpLabel
|
||||
%58 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %58
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %59
|
||||
%62 = OpLabel
|
||||
%65 = OpAccessChain %_ptr_StorageBuffer__arr_mat4v3half_uint_4 %s %uint_0
|
||||
%68 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %u %uint_0
|
||||
%69 = OpLoad %_arr_mat4x3_f16_uint_4 %68
|
||||
%66 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %69
|
||||
OpStore %65 %66
|
||||
%73 = OpAccessChain %_ptr_StorageBuffer_mat4v3half %s %uint_0 %int_1
|
||||
%77 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %u %uint_0 %uint_2
|
||||
%78 = OpLoad %mat4x3_f16 %77
|
||||
%74 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %78
|
||||
OpStore %73 %74
|
||||
%81 = OpAccessChain %_ptr_StorageBuffer_v3half %s %uint_0 %int_1 %79
|
||||
%83 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %33 %uint_1
|
||||
%84 = OpLoad %v3half %83
|
||||
%85 = OpVectorShuffle %v3half %84 %84 2 0 1
|
||||
OpStore %81 %85
|
||||
%87 = OpAccessChain %_ptr_StorageBuffer_half %s %uint_0 %int_1 %79 %uint_0
|
||||
%89 = OpAccessChain %_ptr_Uniform_half %u %uint_0 %33 %uint_1 %33
|
||||
%90 = OpLoad %half %89
|
||||
OpStore %87 %90
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,13 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
@group(0) @binding(1) var<storage, read_write> s : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
s = u;
|
||||
s[1] = u[2];
|
||||
s[1][0] = u[0][1].zxy;
|
||||
s[1][0].x = u[0][1].x;
|
||||
}
|
||||
@@ -0,0 +1,12 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
var<workgroup> w : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
w = u;
|
||||
w[1] = u[2];
|
||||
w[1][0] = u[0][1].zxy;
|
||||
w[1][0].x = u[0][1].x;
|
||||
}
|
||||
@@ -0,0 +1,66 @@
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
groupshared matrix<float16_t, 4, 3> w[4];
|
||||
|
||||
struct tint_symbol_1 {
|
||||
uint local_invocation_index : SV_GroupIndex;
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_3(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_2_ret[4];
|
||||
tint_symbol_2_ret tint_symbol_2(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_3(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f_inner(uint local_invocation_index) {
|
||||
{
|
||||
for(uint idx = local_invocation_index; (idx < 4u); idx = (idx + 1u)) {
|
||||
const uint i = idx;
|
||||
w[i] = matrix<float16_t, 4, 3>((float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx);
|
||||
}
|
||||
}
|
||||
GroupMemoryBarrierWithGroupSync();
|
||||
w = tint_symbol_2(u, 0u);
|
||||
w[1] = tint_symbol_3(u, 64u);
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
w[1][0] = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy;
|
||||
w[1][0].x = float16_t(f16tof32(((u[0].z) & 0xFFFF)));
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f(tint_symbol_1 tint_symbol) {
|
||||
f_inner(tint_symbol.local_invocation_index);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,71 @@
|
||||
SKIP: FAILED
|
||||
|
||||
cbuffer cbuffer_u : register(b0, space0) {
|
||||
uint4 u[8];
|
||||
};
|
||||
groupshared matrix<float16_t, 4, 3> w[4];
|
||||
|
||||
struct tint_symbol_1 {
|
||||
uint local_invocation_index : SV_GroupIndex;
|
||||
};
|
||||
|
||||
matrix<float16_t, 4, 3> tint_symbol_3(uint4 buffer[8], uint offset) {
|
||||
const uint scalar_offset = ((offset + 0u)) / 4;
|
||||
uint4 ubo_load_1 = buffer[scalar_offset / 4];
|
||||
uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
|
||||
vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
|
||||
float16_t ubo_load_y = f16tof32(ubo_load[0] >> 16);
|
||||
const uint scalar_offset_1 = ((offset + 8u)) / 4;
|
||||
uint4 ubo_load_3 = buffer[scalar_offset_1 / 4];
|
||||
uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
|
||||
vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
|
||||
float16_t ubo_load_2_y = f16tof32(ubo_load_2[0] >> 16);
|
||||
const uint scalar_offset_2 = ((offset + 16u)) / 4;
|
||||
uint4 ubo_load_5 = buffer[scalar_offset_2 / 4];
|
||||
uint2 ubo_load_4 = ((scalar_offset_2 & 2) ? ubo_load_5.zw : ubo_load_5.xy);
|
||||
vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
|
||||
float16_t ubo_load_4_y = f16tof32(ubo_load_4[0] >> 16);
|
||||
const uint scalar_offset_3 = ((offset + 24u)) / 4;
|
||||
uint4 ubo_load_7 = buffer[scalar_offset_3 / 4];
|
||||
uint2 ubo_load_6 = ((scalar_offset_3 & 2) ? ubo_load_7.zw : ubo_load_7.xy);
|
||||
vector<float16_t, 2> ubo_load_6_xz = vector<float16_t, 2>(f16tof32(ubo_load_6 & 0xFFFF));
|
||||
float16_t ubo_load_6_y = f16tof32(ubo_load_6[0] >> 16);
|
||||
return matrix<float16_t, 4, 3>(vector<float16_t, 3>(ubo_load_xz[0], ubo_load_y, ubo_load_xz[1]), vector<float16_t, 3>(ubo_load_2_xz[0], ubo_load_2_y, ubo_load_2_xz[1]), vector<float16_t, 3>(ubo_load_4_xz[0], ubo_load_4_y, ubo_load_4_xz[1]), vector<float16_t, 3>(ubo_load_6_xz[0], ubo_load_6_y, ubo_load_6_xz[1]));
|
||||
}
|
||||
|
||||
typedef matrix<float16_t, 4, 3> tint_symbol_2_ret[4];
|
||||
tint_symbol_2_ret tint_symbol_2(uint4 buffer[8], uint offset) {
|
||||
matrix<float16_t, 4, 3> arr[4] = (matrix<float16_t, 4, 3>[4])0;
|
||||
{
|
||||
for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
|
||||
arr[i_1] = tint_symbol_3(buffer, (offset + (i_1 * 32u)));
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f_inner(uint local_invocation_index) {
|
||||
{
|
||||
for(uint idx = local_invocation_index; (idx < 4u); idx = (idx + 1u)) {
|
||||
const uint i = idx;
|
||||
w[i] = matrix<float16_t, 4, 3>((float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx, (float16_t(0.0h)).xxx);
|
||||
}
|
||||
}
|
||||
GroupMemoryBarrierWithGroupSync();
|
||||
w = tint_symbol_2(u, 0u);
|
||||
w[1] = tint_symbol_3(u, 64u);
|
||||
uint2 ubo_load_8 = u[0].zw;
|
||||
vector<float16_t, 2> ubo_load_8_xz = vector<float16_t, 2>(f16tof32(ubo_load_8 & 0xFFFF));
|
||||
float16_t ubo_load_8_y = f16tof32(ubo_load_8[0] >> 16);
|
||||
w[1][0] = vector<float16_t, 3>(ubo_load_8_xz[0], ubo_load_8_y, ubo_load_8_xz[1]).zxy;
|
||||
w[1][0].x = float16_t(f16tof32(((u[0].z) & 0xFFFF)));
|
||||
}
|
||||
|
||||
[numthreads(1, 1, 1)]
|
||||
void f(tint_symbol_1 tint_symbol) {
|
||||
f_inner(tint_symbol.local_invocation_index);
|
||||
return;
|
||||
}
|
||||
FXC validation failure:
|
||||
D:\Projects\RampUp\dawn\test\tint\buffer\Shader@0x000001D6F2EDF360(4,20-28): error X3000: syntax error: unexpected token 'float16_t'
|
||||
|
||||
@@ -0,0 +1,48 @@
|
||||
#version 310 es
|
||||
#extension GL_AMD_gpu_shader_half_float : require
|
||||
|
||||
struct mat4x3_f16 {
|
||||
f16vec3 col0;
|
||||
f16vec3 col1;
|
||||
f16vec3 col2;
|
||||
f16vec3 col3;
|
||||
};
|
||||
|
||||
layout(binding = 0, std140) uniform u_block_std140_ubo {
|
||||
mat4x3_f16 inner[4];
|
||||
} u;
|
||||
|
||||
shared f16mat4x3 w[4];
|
||||
f16mat4x3 conv_mat4x3_f16(mat4x3_f16 val) {
|
||||
return f16mat4x3(val.col0, val.col1, val.col2, val.col3);
|
||||
}
|
||||
|
||||
f16mat4x3[4] conv_arr4_mat4x3_f16(mat4x3_f16 val[4]) {
|
||||
f16mat4x3 arr[4] = f16mat4x3[4](f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf), f16mat4x3(0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf, 0.0hf));
|
||||
{
|
||||
for(uint i = 0u; (i < 4u); i = (i + 1u)) {
|
||||
arr[i] = conv_mat4x3_f16(val[i]);
|
||||
}
|
||||
}
|
||||
return arr;
|
||||
}
|
||||
|
||||
void f(uint local_invocation_index) {
|
||||
{
|
||||
for(uint idx = local_invocation_index; (idx < 4u); idx = (idx + 1u)) {
|
||||
uint i = idx;
|
||||
w[i] = f16mat4x3(f16vec3(0.0hf), f16vec3(0.0hf), f16vec3(0.0hf), f16vec3(0.0hf));
|
||||
}
|
||||
}
|
||||
barrier();
|
||||
w = conv_arr4_mat4x3_f16(u.inner);
|
||||
w[1] = conv_mat4x3_f16(u.inner[2u]);
|
||||
w[1][0] = u.inner[0u].col1.zxy;
|
||||
w[1][0].x = u.inner[0u].col1[0u];
|
||||
}
|
||||
|
||||
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
void main() {
|
||||
f(gl_LocalInvocationIndex);
|
||||
return;
|
||||
}
|
||||
@@ -0,0 +1,38 @@
|
||||
#include <metal_stdlib>
|
||||
|
||||
using namespace metal;
|
||||
|
||||
template<typename T, size_t N>
|
||||
struct tint_array {
|
||||
const constant T& operator[](size_t i) const constant { return elements[i]; }
|
||||
device T& operator[](size_t i) device { return elements[i]; }
|
||||
const device T& operator[](size_t i) const device { return elements[i]; }
|
||||
thread T& operator[](size_t i) thread { return elements[i]; }
|
||||
const thread T& operator[](size_t i) const thread { return elements[i]; }
|
||||
threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
|
||||
const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
|
||||
T elements[N];
|
||||
};
|
||||
|
||||
struct tint_symbol_5 {
|
||||
tint_array<half4x3, 4> w;
|
||||
};
|
||||
|
||||
void f_inner(uint local_invocation_index, threadgroup tint_array<half4x3, 4>* const tint_symbol, const constant tint_array<half4x3, 4>* const tint_symbol_1) {
|
||||
for(uint idx = local_invocation_index; (idx < 4u); idx = (idx + 1u)) {
|
||||
uint const i = idx;
|
||||
(*(tint_symbol))[i] = half4x3(half3(0.0h), half3(0.0h), half3(0.0h), half3(0.0h));
|
||||
}
|
||||
threadgroup_barrier(mem_flags::mem_threadgroup);
|
||||
*(tint_symbol) = *(tint_symbol_1);
|
||||
(*(tint_symbol))[1] = (*(tint_symbol_1))[2];
|
||||
(*(tint_symbol))[1][0] = half3((*(tint_symbol_1))[0][1]).zxy;
|
||||
(*(tint_symbol))[1][0][0] = (*(tint_symbol_1))[0][1][0];
|
||||
}
|
||||
|
||||
kernel void f(const constant tint_array<half4x3, 4>* tint_symbol_4 [[buffer(0)]], threadgroup tint_symbol_5* tint_symbol_3 [[threadgroup(0)]], uint local_invocation_index [[thread_index_in_threadgroup]]) {
|
||||
threadgroup tint_array<half4x3, 4>* const tint_symbol_2 = &((*(tint_symbol_3)).w);
|
||||
f_inner(local_invocation_index, tint_symbol_2, tint_symbol_4);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -0,0 +1,192 @@
|
||||
; SPIR-V
|
||||
; Version: 1.3
|
||||
; Generator: Google Tint Compiler; 0
|
||||
; Bound: 113
|
||||
; Schema: 0
|
||||
OpCapability Shader
|
||||
OpCapability Float16
|
||||
OpCapability UniformAndStorageBuffer16BitAccess
|
||||
OpCapability StorageBuffer16BitAccess
|
||||
OpCapability StorageInputOutput16
|
||||
OpMemoryModel Logical GLSL450
|
||||
OpEntryPoint GLCompute %f "f" %local_invocation_index_1
|
||||
OpExecutionMode %f LocalSize 1 1 1
|
||||
OpName %local_invocation_index_1 "local_invocation_index_1"
|
||||
OpName %u_block_std140 "u_block_std140"
|
||||
OpMemberName %u_block_std140 0 "inner"
|
||||
OpName %mat4x3_f16 "mat4x3_f16"
|
||||
OpMemberName %mat4x3_f16 0 "col0"
|
||||
OpMemberName %mat4x3_f16 1 "col1"
|
||||
OpMemberName %mat4x3_f16 2 "col2"
|
||||
OpMemberName %mat4x3_f16 3 "col3"
|
||||
OpName %u "u"
|
||||
OpName %w "w"
|
||||
OpName %conv_mat4x3_f16 "conv_mat4x3_f16"
|
||||
OpName %val "val"
|
||||
OpName %conv_arr4_mat4x3_f16 "conv_arr4_mat4x3_f16"
|
||||
OpName %val_0 "val"
|
||||
OpName %arr "arr"
|
||||
OpName %i "i"
|
||||
OpName %var_for_index "var_for_index"
|
||||
OpName %f_inner "f_inner"
|
||||
OpName %local_invocation_index "local_invocation_index"
|
||||
OpName %idx "idx"
|
||||
OpName %f "f"
|
||||
OpDecorate %local_invocation_index_1 BuiltIn LocalInvocationIndex
|
||||
OpDecorate %u_block_std140 Block
|
||||
OpMemberDecorate %u_block_std140 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 0 Offset 0
|
||||
OpMemberDecorate %mat4x3_f16 1 Offset 8
|
||||
OpMemberDecorate %mat4x3_f16 2 Offset 16
|
||||
OpMemberDecorate %mat4x3_f16 3 Offset 24
|
||||
OpDecorate %_arr_mat4x3_f16_uint_4 ArrayStride 32
|
||||
OpDecorate %u NonWritable
|
||||
OpDecorate %u DescriptorSet 0
|
||||
OpDecorate %u Binding 0
|
||||
OpDecorate %_arr_mat4v3half_uint_4 ArrayStride 32
|
||||
%uint = OpTypeInt 32 0
|
||||
%_ptr_Input_uint = OpTypePointer Input %uint
|
||||
%local_invocation_index_1 = OpVariable %_ptr_Input_uint Input
|
||||
%half = OpTypeFloat 16
|
||||
%v3half = OpTypeVector %half 3
|
||||
%mat4x3_f16 = OpTypeStruct %v3half %v3half %v3half %v3half
|
||||
%uint_4 = OpConstant %uint 4
|
||||
%_arr_mat4x3_f16_uint_4 = OpTypeArray %mat4x3_f16 %uint_4
|
||||
%u_block_std140 = OpTypeStruct %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Uniform_u_block_std140 = OpTypePointer Uniform %u_block_std140
|
||||
%u = OpVariable %_ptr_Uniform_u_block_std140 Uniform
|
||||
%mat4v3half = OpTypeMatrix %v3half 4
|
||||
%_arr_mat4v3half_uint_4 = OpTypeArray %mat4v3half %uint_4
|
||||
%_ptr_Workgroup__arr_mat4v3half_uint_4 = OpTypePointer Workgroup %_arr_mat4v3half_uint_4
|
||||
%w = OpVariable %_ptr_Workgroup__arr_mat4v3half_uint_4 Workgroup
|
||||
%16 = OpTypeFunction %mat4v3half %mat4x3_f16
|
||||
%25 = OpTypeFunction %_arr_mat4v3half_uint_4 %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function__arr_mat4v3half_uint_4 = OpTypePointer Function %_arr_mat4v3half_uint_4
|
||||
%31 = OpConstantNull %_arr_mat4v3half_uint_4
|
||||
%_ptr_Function_uint = OpTypePointer Function %uint
|
||||
%34 = OpConstantNull %uint
|
||||
%bool = OpTypeBool
|
||||
%_ptr_Function__arr_mat4x3_f16_uint_4 = OpTypePointer Function %_arr_mat4x3_f16_uint_4
|
||||
%47 = OpConstantNull %_arr_mat4x3_f16_uint_4
|
||||
%_ptr_Function_mat4v3half = OpTypePointer Function %mat4v3half
|
||||
%_ptr_Function_mat4x3_f16 = OpTypePointer Function %mat4x3_f16
|
||||
%uint_1 = OpConstant %uint 1
|
||||
%void = OpTypeVoid
|
||||
%60 = OpTypeFunction %void %uint
|
||||
%_ptr_Workgroup_mat4v3half = OpTypePointer Workgroup %mat4v3half
|
||||
%78 = OpConstantNull %mat4v3half
|
||||
%uint_2 = OpConstant %uint 2
|
||||
%uint_264 = OpConstant %uint 264
|
||||
%uint_0 = OpConstant %uint 0
|
||||
%_ptr_Uniform__arr_mat4x3_f16_uint_4 = OpTypePointer Uniform %_arr_mat4x3_f16_uint_4
|
||||
%int = OpTypeInt 32 1
|
||||
%int_1 = OpConstant %int 1
|
||||
%_ptr_Uniform_mat4x3_f16 = OpTypePointer Uniform %mat4x3_f16
|
||||
%96 = OpConstantNull %int
|
||||
%_ptr_Workgroup_v3half = OpTypePointer Workgroup %v3half
|
||||
%_ptr_Uniform_v3half = OpTypePointer Uniform %v3half
|
||||
%_ptr_Workgroup_half = OpTypePointer Workgroup %half
|
||||
%_ptr_Uniform_half = OpTypePointer Uniform %half
|
||||
%108 = OpTypeFunction %void
|
||||
%conv_mat4x3_f16 = OpFunction %mat4v3half None %16
|
||||
%val = OpFunctionParameter %mat4x3_f16
|
||||
%19 = OpLabel
|
||||
%20 = OpCompositeExtract %v3half %val 0
|
||||
%21 = OpCompositeExtract %v3half %val 1
|
||||
%22 = OpCompositeExtract %v3half %val 2
|
||||
%23 = OpCompositeExtract %v3half %val 3
|
||||
%24 = OpCompositeConstruct %mat4v3half %20 %21 %22 %23
|
||||
OpReturnValue %24
|
||||
OpFunctionEnd
|
||||
%conv_arr4_mat4x3_f16 = OpFunction %_arr_mat4v3half_uint_4 None %25
|
||||
%val_0 = OpFunctionParameter %_arr_mat4x3_f16_uint_4
|
||||
%28 = OpLabel
|
||||
%arr = OpVariable %_ptr_Function__arr_mat4v3half_uint_4 Function %31
|
||||
%i = OpVariable %_ptr_Function_uint Function %34
|
||||
%var_for_index = OpVariable %_ptr_Function__arr_mat4x3_f16_uint_4 Function %47
|
||||
OpBranch %35
|
||||
%35 = OpLabel
|
||||
OpLoopMerge %36 %37 None
|
||||
OpBranch %38
|
||||
%38 = OpLabel
|
||||
%40 = OpLoad %uint %i
|
||||
%41 = OpULessThan %bool %40 %uint_4
|
||||
%39 = OpLogicalNot %bool %41
|
||||
OpSelectionMerge %43 None
|
||||
OpBranchConditional %39 %44 %43
|
||||
%44 = OpLabel
|
||||
OpBranch %36
|
||||
%43 = OpLabel
|
||||
OpStore %var_for_index %val_0
|
||||
%48 = OpLoad %uint %i
|
||||
%50 = OpAccessChain %_ptr_Function_mat4v3half %arr %48
|
||||
%52 = OpLoad %uint %i
|
||||
%54 = OpAccessChain %_ptr_Function_mat4x3_f16 %var_for_index %52
|
||||
%55 = OpLoad %mat4x3_f16 %54
|
||||
%51 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %55
|
||||
OpStore %50 %51
|
||||
OpBranch %37
|
||||
%37 = OpLabel
|
||||
%56 = OpLoad %uint %i
|
||||
%58 = OpIAdd %uint %56 %uint_1
|
||||
OpStore %i %58
|
||||
OpBranch %35
|
||||
%36 = OpLabel
|
||||
%59 = OpLoad %_arr_mat4v3half_uint_4 %arr
|
||||
OpReturnValue %59
|
||||
OpFunctionEnd
|
||||
%f_inner = OpFunction %void None %60
|
||||
%local_invocation_index = OpFunctionParameter %uint
|
||||
%64 = OpLabel
|
||||
%idx = OpVariable %_ptr_Function_uint Function %34
|
||||
OpStore %idx %local_invocation_index
|
||||
OpBranch %66
|
||||
%66 = OpLabel
|
||||
OpLoopMerge %67 %68 None
|
||||
OpBranch %69
|
||||
%69 = OpLabel
|
||||
%71 = OpLoad %uint %idx
|
||||
%72 = OpULessThan %bool %71 %uint_4
|
||||
%70 = OpLogicalNot %bool %72
|
||||
OpSelectionMerge %73 None
|
||||
OpBranchConditional %70 %74 %73
|
||||
%74 = OpLabel
|
||||
OpBranch %67
|
||||
%73 = OpLabel
|
||||
%75 = OpLoad %uint %idx
|
||||
%77 = OpAccessChain %_ptr_Workgroup_mat4v3half %w %75
|
||||
OpStore %77 %78
|
||||
OpBranch %68
|
||||
%68 = OpLabel
|
||||
%79 = OpLoad %uint %idx
|
||||
%80 = OpIAdd %uint %79 %uint_1
|
||||
OpStore %idx %80
|
||||
OpBranch %66
|
||||
%67 = OpLabel
|
||||
OpControlBarrier %uint_2 %uint_2 %uint_264
|
||||
%87 = OpAccessChain %_ptr_Uniform__arr_mat4x3_f16_uint_4 %u %uint_0
|
||||
%88 = OpLoad %_arr_mat4x3_f16_uint_4 %87
|
||||
%84 = OpFunctionCall %_arr_mat4v3half_uint_4 %conv_arr4_mat4x3_f16 %88
|
||||
OpStore %w %84
|
||||
%91 = OpAccessChain %_ptr_Workgroup_mat4v3half %w %int_1
|
||||
%94 = OpAccessChain %_ptr_Uniform_mat4x3_f16 %u %uint_0 %uint_2
|
||||
%95 = OpLoad %mat4x3_f16 %94
|
||||
%92 = OpFunctionCall %mat4v3half %conv_mat4x3_f16 %95
|
||||
OpStore %91 %92
|
||||
%98 = OpAccessChain %_ptr_Workgroup_v3half %w %int_1 %96
|
||||
%100 = OpAccessChain %_ptr_Uniform_v3half %u %uint_0 %34 %uint_1
|
||||
%101 = OpLoad %v3half %100
|
||||
%102 = OpVectorShuffle %v3half %101 %101 2 0 1
|
||||
OpStore %98 %102
|
||||
%104 = OpAccessChain %_ptr_Workgroup_half %w %int_1 %96 %uint_0
|
||||
%106 = OpAccessChain %_ptr_Uniform_half %u %uint_0 %34 %uint_1 %34
|
||||
%107 = OpLoad %half %106
|
||||
OpStore %104 %107
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
%f = OpFunction %void None %108
|
||||
%110 = OpLabel
|
||||
%112 = OpLoad %uint %local_invocation_index_1
|
||||
%111 = OpFunctionCall %void %f_inner %112
|
||||
OpReturn
|
||||
OpFunctionEnd
|
||||
@@ -0,0 +1,13 @@
|
||||
enable f16;
|
||||
|
||||
@group(0) @binding(0) var<uniform> u : array<mat4x3<f16>, 4>;
|
||||
|
||||
var<workgroup> w : array<mat4x3<f16>, 4>;
|
||||
|
||||
@compute @workgroup_size(1)
|
||||
fn f() {
|
||||
w = u;
|
||||
w[1] = u[2];
|
||||
w[1][0] = u[0][1].zxy;
|
||||
w[1][0].x = u[0][1].x;
|
||||
}
|
||||
Reference in New Issue
Block a user