ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Hacked in shader disassembler. 

Checking this in before I break it. Has tons of issues.
Turns out the 360 GPU uses an Adreno (rAdeon)-like r2xx shader format.
VC++'s handling of bitfields is absolutely terrible.
This commit is contained in:
Ben Vanik 2013-10-09 19:35:54 -07:00
parent 17be429244
commit 7b97d475c2
6 changed files with 1235 additions and 5629 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/xenia/gpu/nop/nop_graphics_driver.cc
View File

@ -11,6 +11,7 @@
#include <xenia/gpu/gpu-private.h>
#include <xenia/gpu/ucode/ucode_disassembler.h>
#include <xenia/gpu/ucode/ucode_ops.h>
using namespace xe;
@ -53,13 +54,13 @@ void NopGraphicsDriver::SetShader(
uint8_t* p = xe_memory_addr(memory_, address);
uint32_t dw0 = XEGETUINT32BE(p + 0);
uint32_t* dws = (uint32_t*)xe_malloc(length);
for (int n = 0; n < length; n += 4) {
uint32_t dws[512] = {0};
for (uint32_t n = 0; n < length; n += 4) {
dws[n / 4] = XEGETUINT32BE(p + n);
}
UcodeDisassembler disasm;
//
disasm.Disassemble(dws, length / 4, type == XE_GPU_SHADER_TYPE_PIXEL);
}
void NopGraphicsDriver::DrawIndexed(

724
src/xenia/gpu/ucode/ucode_disassembler.cc
View File

@ -9,14 +9,738 @@
#include <xenia/gpu/ucode/ucode_disassembler.h>
#include <xenia/gpu/ucode/ucode_ops.h>
using namespace xe;
using namespace xe::gpu;
using namespace xe::gpu::ucode;
enum shader_t {
SHADER_VERTEX,
SHADER_FRAGMENT,
SHADER_COMPUTE,
};
/* bitmask of debug flags */
enum debug_t {
PRINT_RAW = 0x1, /* dump raw hexdump */
PRINT_VERBOSE = 0x2,
EXPAND_REPEAT = 0x4,
};
int disasm_a2xx(uint32_t *dwords, int sizedwords, int level, enum shader_t type);
void disasm_set_debug(enum debug_t debug);
FILE* disasm_file = NULL;
UcodeDisassembler::UcodeDisassembler() {
}
UcodeDisassembler::~UcodeDisassembler() {
}
void UcodeDisassembler::Disassemble(uint32_t* dwords, int size_dwords, bool pixel) {
disasm_file = fopen("shader.txt", "w");
disasm_set_debug((debug_t)(PRINT_RAW | PRINT_VERBOSE));
disasm_a2xx(dwords, size_dwords, 1, pixel ? SHADER_FRAGMENT : SHADER_VERTEX);
fclose(disasm_file);
}
/*
* Copyright (c) 2012 Rob Clark <robdclark@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
static const char *levels[] = {
"",
"\t",
"\t\t",
"\t\t\t",
"\t\t\t\t",
"\t\t\t\t\t",
"\t\t\t\t\t\t",
"\t\t\t\t\t\t\t",
"\t\t\t\t\t\t\t\t",
"\t\t\t\t\t\t\t\t\t",
"x",
"x",
"x",
"x",
"x",
"x",
};
enum debug_t debug;
/*
* ALU instructions:
*/
static const char chan_names[] = {
'x', 'y', 'z', 'w',
/* these only apply to FETCH dst's: */
'0', '1', '?', '_',
};
static void print_srcreg(uint32_t num, uint32_t type,
uint32_t swiz, uint32_t negate, uint32_t abs)
{
if (negate)
fprintf(disasm_file, "-");
if (abs)
fprintf(disasm_file, "|");
fprintf(disasm_file, "%c%u", type ? 'R' : 'C', num);
if (swiz) {
int i;
fprintf(disasm_file, ".");
for (i = 0; i < 4; i++) {
fprintf(disasm_file, "%c", chan_names[(swiz + i) & 0x3]);
swiz >>= 2;
}
}
if (abs)
fprintf(disasm_file, "|");
}
static void print_dstreg(uint32_t num, uint32_t mask, uint32_t dst_exp)
{
fprintf(disasm_file, "%s%u", dst_exp ? "export" : "R", num);
if (mask != 0xf) {
int i;
fprintf(disasm_file, ".");
for (i = 0; i < 4; i++) {
fprintf(disasm_file, "%c", (mask & 0x1) ? chan_names[i] : '_');
mask >>= 1;
}
}
}
static void print_export_comment(uint32_t num, enum shader_t type)
{
const char *name = NULL;
switch (type) {
case SHADER_VERTEX:
switch (num) {
case 62: name = "gl_Position"; break;
case 63: name = "gl_PointSize"; break;
}
break;
case SHADER_FRAGMENT:
switch (num) {
case 0: name = "gl_FragColor"; break;
}
break;
}
/* if we had a symbol table here, we could look
* up the name of the varying..
*/
if (name) {
fprintf(disasm_file, "\t; %s", name);
}
}
struct {
uint32_t num_srcs;
const char *name;
} vector_instructions[0x20] = {
#define INSTR(opc, num_srcs) { num_srcs, #opc }
INSTR(ADDv, 2), // 0
INSTR(MULv, 2), // 1
INSTR(MAXv, 2), // 2
INSTR(MINv, 2), // 3
INSTR(SETEv, 2), // 4
INSTR(SETGTv, 2), // 5
INSTR(SETGTEv, 2), // 6
INSTR(SETNEv, 2), // 7
INSTR(FRACv, 1), // 8
INSTR(TRUNCv, 1), // 9
INSTR(FLOORv, 1), // 10
INSTR(MULADDv, 3), // 111
INSTR(CNDEv, 3), // 12
INSTR(CNDGTEv, 3), // 13
INSTR(CNDGTv, 3), // 14
INSTR(DOT4v, 2), // 15
INSTR(DOT3v, 2), // 16
INSTR(DOT2ADDv, 3), // 17 -- ???
INSTR(CUBEv, 2), // 18
INSTR(MAX4v, 1), // 19
INSTR(PRED_SETE_PUSHv, 2), // 20
INSTR(PRED_SETNE_PUSHv, 2), // 21
INSTR(PRED_SETGT_PUSHv, 2), // 22
INSTR(PRED_SETGTE_PUSHv, 2), // 23
INSTR(KILLEv, 2), // 24
INSTR(KILLGTv, 2), // 25
INSTR(KILLGTEv, 2), // 26
INSTR(KILLNEv, 2), // 27
INSTR(DSTv, 2), // 28
INSTR(MOVAv, 1), // 29
}, scalar_instructions[0x40] = {
INSTR(ADDs, 1), // 0
INSTR(ADD_PREVs, 1), // 1
INSTR(MULs, 1), // 2
INSTR(MUL_PREVs, 1), // 3
INSTR(MUL_PREV2s, 1), // 4
INSTR(MAXs, 1), // 5
INSTR(MINs, 1), // 6
INSTR(SETEs, 1), // 7
INSTR(SETGTs, 1), // 8
INSTR(SETGTEs, 1), // 9
INSTR(SETNEs, 1), // 10
INSTR(FRACs, 1), // 11
INSTR(TRUNCs, 1), // 12
INSTR(FLOORs, 1), // 13
INSTR(EXP_IEEE, 1), // 14
INSTR(LOG_CLAMP, 1), // 15
INSTR(LOG_IEEE, 1), // 16
INSTR(RECIP_CLAMP, 1), // 17
INSTR(RECIP_FF, 1), // 18
INSTR(RECIP_IEEE, 1), // 19
INSTR(RECIPSQ_CLAMP, 1), // 20
INSTR(RECIPSQ_FF, 1), // 21
INSTR(RECIPSQ_IEEE, 1), // 22
INSTR(MOVAs, 1), // 23
INSTR(MOVA_FLOORs, 1), // 24
INSTR(SUBs, 1), // 25
INSTR(SUB_PREVs, 1), // 26
INSTR(PRED_SETEs, 1), // 27
INSTR(PRED_SETNEs, 1), // 28
INSTR(PRED_SETGTs, 1), // 29
INSTR(PRED_SETGTEs, 1), // 30
INSTR(PRED_SET_INVs, 1), // 31
INSTR(PRED_SET_POPs, 1), // 32
INSTR(PRED_SET_CLRs, 1), // 33
INSTR(PRED_SET_RESTOREs, 1), // 34
INSTR(KILLEs, 1), // 35
INSTR(KILLGTs, 1), // 36
INSTR(KILLGTEs, 1), // 37
INSTR(KILLNEs, 1), // 38
INSTR(KILLONEs, 1), // 39
INSTR(SQRT_IEEE, 1), // 40
{0, 0},
INSTR(MUL_CONST_0, 1), // 42
INSTR(MUL_CONST_1, 1), // 43
INSTR(ADD_CONST_0, 1), // 44
INSTR(ADD_CONST_1, 1), // 45
INSTR(SUB_CONST_0, 1), // 46
INSTR(SUB_CONST_1, 1), // 47
INSTR(SIN, 1), // 48
INSTR(COS, 1), // 49
INSTR(RETAIN_PREV, 1), // 50
#undef INSTR
};
static int disasm_alu(uint32_t *dwords, uint32_t alu_off,
int level, int sync, enum shader_t type)
{
instr_alu_t *alu = (instr_alu_t *)dwords;
fprintf(disasm_file, "%s", levels[level]);
if (debug & PRINT_RAW) {
fprintf(disasm_file, "%02x: %08x %08x %08x\t", alu_off,
dwords[0], dwords[1], dwords[2]);
}
fprintf(disasm_file, " %sALU:\t", sync ? "(S)" : " ");
fprintf(disasm_file, "%s", vector_instructions[alu->vector_opc].name);
if (alu->pred_select & 0x2) {
/* seems to work similar to conditional execution in ARM instruction
* set, so let's use a similar syntax for now:
*/
fprintf(disasm_file, (alu->pred_select & 0x1) ? "EQ" : "NE");
}
fprintf(disasm_file, "\t");
print_dstreg(alu->vector_dest, alu->vector_write_mask, alu->export_data);
fprintf(disasm_file, " = ");
if (vector_instructions[alu->vector_opc].num_srcs == 3) {
print_srcreg(alu->src3_reg, alu->src3_sel, alu->src3_swiz,
alu->src3_reg_negate, alu->src3_reg_abs);
fprintf(disasm_file, ", ");
}
print_srcreg(alu->src1_reg, alu->src1_sel, alu->src1_swiz,
alu->src1_reg_negate, alu->src1_reg_abs);
if (vector_instructions[alu->vector_opc].num_srcs > 1) {
fprintf(disasm_file, ", ");
print_srcreg(alu->src2_reg, alu->src2_sel, alu->src2_swiz,
alu->src2_reg_negate, alu->src2_reg_abs);
}
if (alu->vector_clamp)
fprintf(disasm_file, " CLAMP");
if (alu->export_data)
print_export_comment(alu->vector_dest, type);
fprintf(disasm_file, "\n");
if (alu->scalar_write_mask || !alu->vector_write_mask) {
/* 2nd optional scalar op: */
fprintf(disasm_file, "%s", levels[level]);
if (debug & PRINT_RAW)
fprintf(disasm_file, " \t");
if (scalar_instructions[alu->scalar_opc].name) {
fprintf(disasm_file, "\t \t%s\t", scalar_instructions[alu->scalar_opc].name);
} else {
fprintf(disasm_file, "\t \tOP(%u)\t", alu->scalar_opc);
}
print_dstreg(alu->scalar_dest, alu->scalar_write_mask, alu->export_data);
fprintf(disasm_file, " = ");
print_srcreg(alu->src3_reg, alu->src3_sel, alu->src3_swiz,
alu->src3_reg_negate, alu->src3_reg_abs);
// TODO ADD/MUL must have another src?!?
if (alu->scalar_clamp)
fprintf(disasm_file, " CLAMP");
if (alu->export_data)
print_export_comment(alu->scalar_dest, type);
fprintf(disasm_file, "\n");
}
return 0;
}
/*
* FETCH instructions:
*/
struct {
const char *name;
} fetch_types[0xff] = {
#define TYPE(id) { #id }
TYPE(FMT_1_REVERSE), // 0
{0},
TYPE(FMT_8), // 2
{0},
{0},
{0},
TYPE(FMT_8_8_8_8), // 6
{0},
{0},
{0},
TYPE(FMT_8_8), // 10
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
TYPE(FMT_16), // 24
TYPE(FMT_16_16), // 25
TYPE(FMT_16_16_16_16), // 26
{0},
{0},
{0},
{0},
{0},
{0},
TYPE(FMT_32), // 33
TYPE(FMT_32_32), // 34
TYPE(FMT_32_32_32_32), // 35
TYPE(FMT_32_FLOAT), // 36
TYPE(FMT_32_32_FLOAT), // 37
TYPE(FMT_32_32_32_32_FLOAT), // 38
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
{0},
TYPE(FMT_32_32_32_FLOAT), // 57
#undef TYPE
};
static void print_fetch_dst(uint32_t dst_reg, uint32_t dst_swiz)
{
int i;
fprintf(disasm_file, "\tR%u.", dst_reg);
for (i = 0; i < 4; i++) {
fprintf(disasm_file, "%c", chan_names[dst_swiz & 0x7]);
dst_swiz >>= 3;
}
}
static void print_fetch_vtx(instr_fetch_t *fetch)
{
instr_fetch_vtx_t *vtx = &fetch->vtx;
if (vtx->pred_select) {
/* seems to work similar to conditional execution in ARM instruction
* set, so let's use a similar syntax for now:
*/
fprintf(disasm_file, vtx->pred_condition ? "EQ" : "NE");
}
print_fetch_dst(vtx->dst_reg, vtx->dst_swiz);
fprintf(disasm_file, " = R%u.", vtx->src_reg);
fprintf(disasm_file, "%c", chan_names[vtx->src_swiz & 0x3]);
if (fetch_types[vtx->format].name) {
fprintf(disasm_file, " %s", fetch_types[vtx->format].name);
} else {
fprintf(disasm_file, " TYPE(0x%x)", vtx->format);
}
fprintf(disasm_file, " %s", vtx->format_comp_all ? "SIGNED" : "UNSIGNED");
if (!vtx->num_format_all)
fprintf(disasm_file, " NORMALIZED");
fprintf(disasm_file, " STRIDE(%u)", vtx->stride);
if (vtx->offset)
fprintf(disasm_file, " OFFSET(%u)", vtx->offset);
fprintf(disasm_file, " CONST(%u, %u)", vtx->const_index, vtx->const_index_sel);
if (0) {
// XXX
fprintf(disasm_file, " src_reg_am=%u", vtx->src_reg_am);
fprintf(disasm_file, " dst_reg_am=%u", vtx->dst_reg_am);
fprintf(disasm_file, " num_format_all=%u", vtx->num_format_all);
fprintf(disasm_file, " signed_rf_mode_all=%u", vtx->signed_rf_mode_all);
fprintf(disasm_file, " exp_adjust_all=%u", vtx->exp_adjust_all);
}
}
static void print_fetch_tex(instr_fetch_t *fetch)
{
static const char *filter[] = {
"POINT", // TEX_FILTER_POINT
"LINEAR", // TEX_FILTER_LINEAR
"BASEMAP", // TEX_FILTER_BASEMAP
};
static const char *aniso_filter[] = {
"DISABLED", // ANISO_FILTER_DISABLED
"MAX_1_1", // ANISO_FILTER_MAX_1_1
"MAX_2_1", // ANISO_FILTER_MAX_2_1
"MAX_4_1", // ANISO_FILTER_MAX_4_1
"MAX_8_1", // ANISO_FILTER_MAX_8_1
"MAX_16_1", // ANISO_FILTER_MAX_16_1
};
static const char *arbitrary_filter[] = {
"2x4_SYM", // ARBITRARY_FILTER_2X4_SYM
"2x4_ASYM", // ARBITRARY_FILTER_2X4_ASYM
"4x2_SYM", // ARBITRARY_FILTER_4X2_SYM
"4x2_ASYM", // ARBITRARY_FILTER_4X2_ASYM
"4x4_SYM", // ARBITRARY_FILTER_4X4_SYM
"4x4_ASYM", // ARBITRARY_FILTER_4X4_ASYM
};
static const char *sample_loc[] = {
"CENTROID", // SAMPLE_CENTROID
"CENTER", // SAMPLE_CENTER
};
instr_fetch_tex_t *tex = &fetch->tex;
uint32_t src_swiz = tex->src_swiz;
int i;
if (tex->pred_select) {
/* seems to work similar to conditional execution in ARM instruction
* set, so let's use a similar syntax for now:
*/
fprintf(disasm_file, tex->pred_condition ? "EQ" : "NE");
}
print_fetch_dst(tex->dst_reg, tex->dst_swiz);
fprintf(disasm_file, " = R%u.", tex->src_reg);
for (i = 0; i < 3; i++) {
fprintf(disasm_file, "%c", chan_names[src_swiz & 0x3]);
src_swiz >>= 2;
}
fprintf(disasm_file, " CONST(%u)", tex->const_idx);
if (tex->fetch_valid_only)
fprintf(disasm_file, " VALID_ONLY");
if (tex->tx_coord_denorm)
fprintf(disasm_file, " DENORM");
if (tex->mag_filter != TEX_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " MAG(%s)", filter[tex->mag_filter]);
if (tex->min_filter != TEX_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " MIN(%s)", filter[tex->min_filter]);
if (tex->mip_filter != TEX_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " MIP(%s)", filter[tex->mip_filter]);
if (tex->aniso_filter != ANISO_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " ANISO(%s)", aniso_filter[tex->aniso_filter]);
if (tex->arbitrary_filter != ARBITRARY_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " ARBITRARY(%s)", arbitrary_filter[tex->arbitrary_filter]);
if (tex->vol_mag_filter != TEX_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " VOL_MAG(%s)", filter[tex->vol_mag_filter]);
if (tex->vol_min_filter != TEX_FILTER_USE_FETCH_CONST)
fprintf(disasm_file, " VOL_MIN(%s)", filter[tex->vol_min_filter]);
if (!tex->use_comp_lod) {
fprintf(disasm_file, " LOD(%u)", tex->use_comp_lod);
fprintf(disasm_file, " LOD_BIAS(%u)", tex->lod_bias);
}
if (tex->use_reg_lod) {
fprintf(disasm_file, " REG_LOD(%u)", tex->use_reg_lod);
}
if (tex->use_reg_gradients)
fprintf(disasm_file, " USE_REG_GRADIENTS");
fprintf(disasm_file, " LOCATION(%s)", sample_loc[tex->sample_location]);
if (tex->offset_x || tex->offset_y || tex->offset_z)
fprintf(disasm_file, " OFFSET(%u,%u,%u)", tex->offset_x, tex->offset_y, tex->offset_z);
}
struct {
const char *name;
void (*fxn)(instr_fetch_t *cf);
} fetch_instructions[] = {
#define INSTR(opc, name, fxn) { name, fxn }
INSTR(VTX_FETCH, "VERTEX", print_fetch_vtx), // 0
INSTR(TEX_FETCH, "SAMPLE", print_fetch_tex), // 1
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
{0, 0},
INSTR(TEX_GET_BORDER_COLOR_FRAC, "?", print_fetch_tex), // 16
INSTR(TEX_GET_COMP_TEX_LOD, "?", print_fetch_tex), // 17
INSTR(TEX_GET_GRADIENTS, "?", print_fetch_tex), // 18
INSTR(TEX_GET_WEIGHTS, "?", print_fetch_tex), // 19
{0, 0},
{0, 0},
{0, 0},
{0, 0},
INSTR(TEX_SET_TEX_LOD, "SET_TEX_LOD", print_fetch_tex), // 24
INSTR(TEX_SET_GRADIENTS_H, "?", print_fetch_tex), // 25
INSTR(TEX_SET_GRADIENTS_V, "?", print_fetch_tex), // 26
INSTR(TEX_RESERVED_4, "?", print_fetch_tex), // 27
#undef INSTR
};
static int disasm_fetch(uint32_t *dwords, uint32_t alu_off, int level, int sync)
{
instr_fetch_t *fetch = (instr_fetch_t *)dwords;
fprintf(disasm_file, "%s", levels[level]);
if (debug & PRINT_RAW) {
fprintf(disasm_file, "%02x: %08x %08x %08x\t", alu_off,
dwords[0], dwords[1], dwords[2]);
}
fprintf(disasm_file, " %sFETCH:\t", sync ? "(S)" : " ");
fprintf(disasm_file, "%s", fetch_instructions[fetch->opc].name);
fetch_instructions[fetch->opc].fxn(fetch);
fprintf(disasm_file, "\n");
return 0;
}
/*
* CF instructions:
*/
static int cf_exec(instr_cf_t *cf)
{
return (cf->opc == EXEC) ||
(cf->opc == EXEC_END) ||
(cf->opc == COND_EXEC) ||
(cf->opc == COND_EXEC_END) ||
(cf->opc == COND_PRED_EXEC) ||
(cf->opc == COND_PRED_EXEC_END) ||
(cf->opc == COND_EXEC_PRED_CLEAN) ||
(cf->opc == COND_EXEC_PRED_CLEAN_END);
}
static int cf_cond_exec(instr_cf_t *cf)
{
return (cf->opc == COND_EXEC) ||
(cf->opc == COND_EXEC_END) ||
(cf->opc == COND_PRED_EXEC) ||
(cf->opc == COND_PRED_EXEC_END) ||
(cf->opc == COND_EXEC_PRED_CLEAN) ||
(cf->opc == COND_EXEC_PRED_CLEAN_END);
}
static void print_cf_nop(instr_cf_t *cf)
{
}
static void print_cf_exec(instr_cf_t *cf)
{
fprintf(disasm_file, " ADDR(0x%x) CNT(0x%x)", cf->exec.address, cf->exec.count);
if (cf->exec.yeild)
fprintf(disasm_file, " YIELD");
uint8_t vc = cf->exec.vc_hi | (cf->exec.vc_lo << 2);
if (vc)
fprintf(disasm_file, " VC(0x%x)", vc);
if (cf->exec.bool_addr)
fprintf(disasm_file, " BOOL_ADDR(0x%x)", cf->exec.bool_addr);
if (cf->exec.address_mode == ABSOLUTE_ADDR)
fprintf(disasm_file, " ABSOLUTE_ADDR");
if (cf_cond_exec(cf))
fprintf(disasm_file, " COND(%d)", cf->exec.condition);
}
static void print_cf_loop(instr_cf_t *cf)
{
fprintf(disasm_file, " ADDR(0x%x) LOOP_ID(%d)", cf->loop.address, cf->loop.loop_id);
if (cf->loop.address_mode == ABSOLUTE_ADDR)
fprintf(disasm_file, " ABSOLUTE_ADDR");
}
static void print_cf_jmp_call(instr_cf_t *cf)
{
fprintf(disasm_file, " ADDR(0x%x) DIR(%d)", cf->jmp_call.address, cf->jmp_call.direction);
if (cf->jmp_call.force_call)
fprintf(disasm_file, " FORCE_CALL");
if (cf->jmp_call.predicated_jmp)
fprintf(disasm_file, " COND(%d)", cf->jmp_call.condition);
if (cf->jmp_call.bool_addr)
fprintf(disasm_file, " BOOL_ADDR(0x%x)", cf->jmp_call.bool_addr);
if (cf->jmp_call.address_mode == ABSOLUTE_ADDR)
fprintf(disasm_file, " ABSOLUTE_ADDR");
}
static void print_cf_alloc(instr_cf_t *cf)
{
static const char *bufname[] = {
"NO ALLOC", // SQ_NO_ALLOC
"POSITION", // SQ_POSITION
"PARAM/PIXEL", // SQ_PARAMETER_PIXEL
"MEMORY", // SQ_MEMORY
};
fprintf(disasm_file, " %s SIZE(0x%x)", bufname[cf->alloc.buffer_select], cf->alloc.size);
if (cf->alloc.no_serial)
fprintf(disasm_file, " NO_SERIAL");
if (cf->alloc.alloc_mode) // ???
fprintf(disasm_file, " ALLOC_MODE");
}
struct {
const char *name;
void (*fxn)(instr_cf_t *cf);
} cf_instructions[] = {
#define INSTR(opc, fxn) { #opc, fxn }
INSTR(NOP, print_cf_nop),
INSTR(EXEC, print_cf_exec),
INSTR(EXEC_END, print_cf_exec),
INSTR(COND_EXEC, print_cf_exec),
INSTR(COND_EXEC_END, print_cf_exec),
INSTR(COND_PRED_EXEC, print_cf_exec),
INSTR(COND_PRED_EXEC_END, print_cf_exec),
INSTR(LOOP_START, print_cf_loop),
INSTR(LOOP_END, print_cf_loop),
INSTR(COND_CALL, print_cf_jmp_call),
INSTR(RETURN, print_cf_jmp_call),
INSTR(COND_JMP, print_cf_jmp_call),
INSTR(ALLOC, print_cf_alloc),
INSTR(COND_EXEC_PRED_CLEAN, print_cf_exec),
INSTR(COND_EXEC_PRED_CLEAN_END, print_cf_exec),
INSTR(MARK_VS_FETCH_DONE, print_cf_nop), // ??
#undef INSTR
};
static void print_cf(instr_cf_t *cf, int level)
{
fprintf(disasm_file, "%s", levels[level]);
if (debug & PRINT_RAW) {
uint16_t *words = (uint16_t *)cf;
fprintf(disasm_file, " %04x %04x %04x \t",
words[0], words[1], words[2]);
}
fprintf(disasm_file, "%s", cf_instructions[cf->opc].name);
cf_instructions[cf->opc].fxn(cf);
fprintf(disasm_file, "\n");
}
/*
* The adreno shader microcode consists of two parts:
* 1) A CF (control-flow) program, at the header of the compiled shader,
* which refers to ALU/FETCH instructions that follow it by address.
* 2) ALU and FETCH instructions
*/
int disasm_a2xx(uint32_t *dwords, int sizedwords, int level, enum shader_t type)
{
instr_cf_t *cfs = (instr_cf_t *)dwords;
int idx, max_idx;
for (idx = 0; ; idx++) {
if (idx > sizedwords) {
// Failed to find a control instruction.
return 1;
}
instr_cf_t *cf = &cfs[idx];
if (cf_exec(cf)) {
max_idx = 2 * cf->exec.address;
break;
}
}
for (idx = 0; idx < max_idx; idx++) {
instr_cf_t *cf = &cfs[idx];
print_cf(cf, level);
if (cf_exec(cf)) {
uint32_t sequence = cf->exec.serialize;
uint32_t i;
for (i = 0; i < cf->exec.count; i++) {
uint32_t alu_off = (cf->exec.address + i);
if (sequence & 0x1) {
disasm_fetch(dwords + alu_off * 3, alu_off, level, sequence & 0x2);
} else {
disasm_alu(dwords + alu_off * 3, alu_off, level, sequence & 0x2, type);
}
sequence >>= 2;
}
}
}
return 0;
}
void disasm_set_debug(enum debug_t d)
{
debug = d;
}

2
src/xenia/gpu/ucode/ucode_disassembler.h
View File

@ -25,6 +25,8 @@ public:
UcodeDisassembler();
~UcodeDisassembler();
void Disassemble(uint32_t* dwords, int size_dwords, bool pixel);
private:
};

5603
src/xenia/gpu/ucode/ucode_ops.cc
View File

File diff suppressed because it is too large Load Diff

510
src/xenia/gpu/ucode/ucode_ops.h
View File

@ -20,53 +20,495 @@ namespace gpu {
namespace ucode {
// This code comes from libxemit by GliGli.
// This code comes from the freedreno project:
// https://github.com/freedreno/freedreno/blob/master/includes/instr-a2xx.h
/*
* Copyright (c) 2012 Rob Clark <robdclark@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
enum a2xx_sq_surfaceformat {
FMT_1_REVERSE = 0,
FMT_1 = 1,
FMT_8 = 2,
FMT_1_5_5_5 = 3,
FMT_5_6_5 = 4,
FMT_6_5_5 = 5,
FMT_8_8_8_8 = 6,
FMT_2_10_10_10 = 7,
FMT_8_A = 8,
FMT_8_B = 9,
FMT_8_8 = 10,
FMT_Cr_Y1_Cb_Y0 = 11,
FMT_Y1_Cr_Y0_Cb = 12,
FMT_5_5_5_1 = 13,
FMT_8_8_8_8_A = 14,
FMT_4_4_4_4 = 15,
FMT_10_11_11 = 16,
FMT_11_11_10 = 17,
FMT_DXT1 = 18,
FMT_DXT2_3 = 19,
FMT_DXT4_5 = 20,
FMT_24_8 = 22,
FMT_24_8_FLOAT = 23,
FMT_16 = 24,
FMT_16_16 = 25,
FMT_16_16_16_16 = 26,
FMT_16_EXPAND = 27,
FMT_16_16_EXPAND = 28,
FMT_16_16_16_16_EXPAND = 29,
FMT_16_FLOAT = 30,
FMT_16_16_FLOAT = 31,
FMT_16_16_16_16_FLOAT = 32,
FMT_32 = 33,
FMT_32_32 = 34,
FMT_32_32_32_32 = 35,
FMT_32_FLOAT = 36,
FMT_32_32_FLOAT = 37,
FMT_32_32_32_32_FLOAT = 38,
FMT_32_AS_8 = 39,
FMT_32_AS_8_8 = 40,
FMT_16_MPEG = 41,
FMT_16_16_MPEG = 42,
FMT_8_INTERLACED = 43,
FMT_32_AS_8_INTERLACED = 44,
FMT_32_AS_8_8_INTERLACED = 45,
FMT_16_INTERLACED = 46,
FMT_16_MPEG_INTERLACED = 47,
FMT_16_16_MPEG_INTERLACED = 48,
FMT_DXN = 49,
FMT_8_8_8_8_AS_16_16_16_16 = 50,
FMT_DXT1_AS_16_16_16_16 = 51,
FMT_DXT2_3_AS_16_16_16_16 = 52,
FMT_DXT4_5_AS_16_16_16_16 = 53,
FMT_2_10_10_10_AS_16_16_16_16 = 54,
FMT_10_11_11_AS_16_16_16_16 = 55,
FMT_11_11_10_AS_16_16_16_16 = 56,
FMT_32_32_32_FLOAT = 57,
FMT_DXT3A = 58,
FMT_DXT5A = 59,
FMT_CTX1 = 60,
FMT_DXT3A_AS_1_1_1_1 = 61,
};
/*
* ALU instructions:
*/
typedef enum {
XEMR_NONE = -1,
XEMR_TEMP = 0,
XEMR_CONST = 1,
XEMR_COLOR_OUT = 2,
XEMR_TEX_FETCH = 3,
} ucode_reg_type_t;
ADDs = 0,
ADD_PREVs = 1,
MULs = 2,
MUL_PREVs = 3,
MUL_PREV2s = 4,
MAXs = 5,
MINs = 6,
SETEs = 7,
SETGTs = 8,
SETGTEs = 9,
SETNEs = 10,
FRACs = 11,
TRUNCs = 12,
FLOORs = 13,
EXP_IEEE = 14,
LOG_CLAMP = 15,
LOG_IEEE = 16,
RECIP_CLAMP = 17,
RECIP_FF = 18,
RECIP_IEEE = 19,
RECIPSQ_CLAMP = 20,
RECIPSQ_FF = 21,
RECIPSQ_IEEE = 22,
MOVAs = 23,
MOVA_FLOORs = 24,
SUBs = 25,
SUB_PREVs = 26,
PRED_SETEs = 27,
PRED_SETNEs = 28,
PRED_SETGTs = 29,
PRED_SETGTEs = 30,
PRED_SET_INVs = 31,
PRED_SET_POPs = 32,
PRED_SET_CLRs = 33,
PRED_SET_RESTOREs = 34,
KILLEs = 35,
KILLGTs = 36,
KILLGTEs = 37,
KILLNEs = 38,
KILLONEs = 39,
SQRT_IEEE = 40,
MUL_CONST_0 = 42,
MUL_CONST_1 = 43,
ADD_CONST_0 = 44,
ADD_CONST_1 = 45,
SUB_CONST_0 = 46,
SUB_CONST_1 = 47,
SIN = 48,
COS = 49,
RETAIN_PREV = 50,
} instr_scalar_opc_t;
typedef enum {
XEMO_NONE = -1,
XEMO_SEQUENCER = 0,
XEMO_ALU_VECTOR = 1,
XEMO_ALU_VECTOR_SAT = 2,
XEMO_ALU_SCALAR = 3,
XEMO_ALU_SCALAR_SAT = 4,
XEMO_FETCHES = 5,
} ucode_op_type_t;
ADDv = 0,
MULv = 1,
MAXv = 2,
MINv = 3,
SETEv = 4,
SETGTv = 5,
SETGTEv = 6,
SETNEv = 7,
FRACv = 8,
TRUNCv = 9,
FLOORv = 10,
MULADDv = 11,
CNDEv = 12,
CNDGTEv = 13,
CNDGTv = 14,
DOT4v = 15,
DOT3v = 16,
DOT2ADDv = 17,
CUBEv = 18,
MAX4v = 19,
PRED_SETE_PUSHv = 20,
PRED_SETNE_PUSHv = 21,
PRED_SETGT_PUSHv = 22,
PRED_SETGTE_PUSHv = 23,
KILLEv = 24,
KILLGTv = 25,
KILLGTEv = 26,
KILLNEv = 27,
DSTv = 28,
MOVAv = 29,
} instr_vector_opc_t;
XEPACKEDSTRUCT(instr_alu_t, {
/* dword0: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t vector_dest : 6;
uint32_t vector_dest_rel : 1;
uint32_t low_precision_16b_fp : 1;
uint32_t scalar_dest : 6;
uint32_t scalar_dest_rel : 1;
uint32_t export_data : 1;
uint32_t vector_write_mask : 4;
uint32_t scalar_write_mask : 4;
uint32_t vector_clamp : 1;
uint32_t scalar_clamp : 1;
uint32_t scalar_opc : 6; // instr_scalar_opc_t
});
/* dword1: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t src3_swiz : 8;
uint32_t src2_swiz : 8;
uint32_t src1_swiz : 8;
uint32_t src3_reg_negate : 1;
uint32_t src2_reg_negate : 1;
uint32_t src1_reg_negate : 1;
uint32_t pred_select : 2;
uint32_t relative_addr : 1;
uint32_t const_1_rel_abs : 1;
uint32_t const_0_rel_abs : 1;
});
/* dword2: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t src3_reg : 6;
uint32_t src3_reg_select : 1;
uint32_t src3_reg_abs : 1;
uint32_t src2_reg : 6;
uint32_t src2_reg_select : 1;
uint32_t src2_reg_abs : 1;
uint32_t src1_reg : 6;
uint32_t src1_reg_select : 1;
uint32_t src1_reg_abs : 1;
uint32_t vector_opc : 5; // instr_vector_opc_t
uint32_t src3_sel : 1;
uint32_t src2_sel : 1;
uint32_t src1_sel : 1;
});
});
typedef struct {
int8_t start;
int8_t end;
} ucode_mask_t;
/*
* CF instructions:
*/
typedef enum {
NOP = 0,
EXEC = 1,
EXEC_END = 2,
COND_EXEC = 3,
COND_EXEC_END = 4,
COND_PRED_EXEC = 5,
COND_PRED_EXEC_END = 6,
LOOP_START = 7,
LOOP_END = 8,
COND_CALL = 9,
RETURN = 10,
COND_JMP = 11,
ALLOC = 12,
COND_EXEC_PRED_CLEAN = 13,
COND_EXEC_PRED_CLEAN_END = 14,
MARK_VS_FETCH_DONE = 15,
} instr_cf_opc_t;
typedef enum {
RELATIVE_ADDR = 0,
ABSOLUTE_ADDR = 1,
} instr_addr_mode_t;
typedef enum {
SQ_NO_ALLOC = 0,
SQ_POSITION = 1,
SQ_PARAMETER_PIXEL = 2,
SQ_MEMORY = 3,
} instr_alloc_type_t;
XEPACKEDSTRUCT(instr_cf_exec_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t address : 12;
uint32_t count : 3;
uint32_t yeild : 1;
uint32_t serialize : 12;
uint32_t vc_hi : 4;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t vc_lo : 2; /* vertex cache? */
uint32_t bool_addr : 8;
uint32_t condition : 1;
uint32_t address_mode : 1; // instr_addr_mode_t
uint32_t opc : 4; // instr_cf_opc_t
});
});
XEPACKEDSTRUCT(instr_cf_loop_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t address : 13;
uint32_t repeat : 1;
uint32_t reserved0 : 2;
uint32_t loop_id : 5;
uint32_t pred_break : 1;
uint32_t reserved1_hi : 10;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t reserved1_lo : 10;
uint32_t condition : 1;
uint32_t address_mode : 1; // instr_addr_mode_t
uint32_t opc : 4; // instr_cf_opc_t
});
});
XEPACKEDSTRUCT(instr_cf_jmp_call_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t address : 13;
uint32_t force_call : 1;
uint32_t predicated_jmp : 1;
uint32_t reserved1_hi : 17;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t reserved1_lo : 1;
uint32_t direction : 1;
uint32_t bool_addr : 8;
uint32_t condition : 1;
uint32_t address_mode : 1; // instr_addr_mode_t
uint32_t opc : 4; // instr_cf_opc_t
});
});
XEPACKEDSTRUCT(instr_cf_alloc_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t size : 3;
uint32_t reserved0_hi : 29;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t reserved0_lo : 8;
uint32_t no_serial : 1;
uint32_t buffer_select : 2; // instr_alloc_type_t
uint32_t alloc_mode : 1;
uint32_t opc : 4; // instr_cf_opc_t
});
});
XEPACKEDUNION(instr_cf_t, {
instr_cf_exec_t exec;
instr_cf_loop_t loop;
instr_cf_jmp_call_t jmp_call;
instr_cf_alloc_t alloc;
XEPACKEDSTRUCTANONYMOUS({
uint32_t : 32;
uint32_t : 12;
uint32_t opc : 4; // instr_cf_opc_t
});
});
typedef struct {
ucode_reg_type_t reg_type;
ucode_mask_t reg_mask[2]; // there can be 2 reg masks
uint8_t swizzle_count;
ucode_mask_t swizzle_mask[2]; // there can be 2 swizzle masks
} ucode_reg_t;
/*
* FETCH instructions:
*/
typedef struct {
const char* name;
ucode_op_type_t op_type;
uint8_t base_bin[12];
ucode_mask_t out_mask;
ucode_reg_t regs[4];
} ucode_op_t;
typedef enum {
VTX_FETCH = 0,
TEX_FETCH = 1,
TEX_GET_BORDER_COLOR_FRAC = 16,
TEX_GET_COMP_TEX_LOD = 17,
TEX_GET_GRADIENTS = 18,
TEX_GET_WEIGHTS = 19,
TEX_SET_TEX_LOD = 24,
TEX_SET_GRADIENTS_H = 25,
TEX_SET_GRADIENTS_V = 26,
TEX_RESERVED_4 = 27,
} instr_fetch_opc_t;
typedef enum {
TEX_FILTER_POINT = 0,
TEX_FILTER_LINEAR = 1,
TEX_FILTER_BASEMAP = 2, /* only applicable for mip-filter */
TEX_FILTER_USE_FETCH_CONST = 3,
} instr_tex_filter_t;
extern const ucode_op_t ucode_ops[];
typedef enum {
ANISO_FILTER_DISABLED = 0,
ANISO_FILTER_MAX_1_1 = 1,
ANISO_FILTER_MAX_2_1 = 2,
ANISO_FILTER_MAX_4_1 = 3,
ANISO_FILTER_MAX_8_1 = 4,
ANISO_FILTER_MAX_16_1 = 5,
ANISO_FILTER_USE_FETCH_CONST = 7,
} instr_aniso_filter_t;
typedef enum {
ARBITRARY_FILTER_2X4_SYM = 0,
ARBITRARY_FILTER_2X4_ASYM = 1,
ARBITRARY_FILTER_4X2_SYM = 2,
ARBITRARY_FILTER_4X2_ASYM = 3,
ARBITRARY_FILTER_4X4_SYM = 4,
ARBITRARY_FILTER_4X4_ASYM = 5,
ARBITRARY_FILTER_USE_FETCH_CONST = 7,
} instr_arbitrary_filter_t;
typedef enum {
SAMPLE_CENTROID = 0,
SAMPLE_CENTER = 1,
} instr_sample_loc_t;
typedef enum a2xx_sq_surfaceformat instr_surf_fmt_t;
XEPACKEDSTRUCT(instr_fetch_tex_t, {
/* dword0: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t opc : 5; // instr_fetch_opc_t
uint32_t src_reg : 6;
uint32_t src_reg_am : 1;
uint32_t dst_reg : 6;
uint32_t dst_reg_am : 1;
uint32_t fetch_valid_only : 1;
uint32_t const_idx : 5;
uint32_t tx_coord_denorm : 1;
uint32_t src_swiz : 6;
});
/* dword1: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t dst_swiz : 12;
uint32_t mag_filter : 2; // instr_tex_filter_t
uint32_t min_filter : 2; // instr_tex_filter_t
uint32_t mip_filter : 2; // instr_tex_filter_t
uint32_t aniso_filter : 3; // instr_aniso_filter_t
uint32_t arbitrary_filter : 3; // instr_arbitrary_filter_t
uint32_t vol_mag_filter : 2; // instr_tex_filter_t
uint32_t vol_min_filter : 2; // instr_tex_filter_t
uint32_t use_comp_lod : 1;
uint32_t use_reg_lod : 2;
uint32_t pred_select : 1;
});
/* dword2: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t use_reg_gradients : 1;
uint32_t sample_location : 1; // instr_sample_loc_t
uint32_t lod_bias : 7;
uint32_t unused : 5;
uint32_t dimension : 2;
uint32_t offset_x : 5;
uint32_t offset_y : 5;
uint32_t offset_z : 5;
uint32_t pred_condition : 1;
});
});
XEPACKEDSTRUCT(instr_fetch_vtx_t, {
/* dword0: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t opc : 5; // instr_fetch_opc_t
uint32_t src_reg : 6;
uint32_t src_reg_am : 1;
uint32_t dst_reg : 6;
uint32_t dst_reg_am : 1;
uint32_t must_be_one : 1;
uint32_t const_index : 5;
uint32_t const_index_sel : 2;
uint32_t reserved0 : 3;
uint32_t src_swiz : 2;
});
/* dword1: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t dst_swiz : 12;
uint32_t format_comp_all : 1; /* '1' for signed, '0' for unsigned? */
uint32_t num_format_all : 1; /* '0' for normalized, '1' for unnormalized */
uint32_t signed_rf_mode_all : 1;
uint32_t reserved1 : 1;
uint32_t format : 6; // instr_surf_fmt_t
uint32_t reserved2 : 1;
uint32_t exp_adjust_all : 7;
uint32_t reserved3 : 1;
uint32_t pred_select : 1;
});
/* dword2: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t stride : 8;
uint32_t offset : 23;
uint32_t pred_condition : 1;
});
});
XEPACKEDUNION(instr_fetch_t, {
instr_fetch_tex_t tex;
instr_fetch_vtx_t vtx;
XEPACKEDSTRUCTANONYMOUS({
/* dword0: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t opc : 5; // instr_fetch_opc_t
uint32_t : 27;
});
/* dword1: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t : 32;
});
/* dword2: */
XEPACKEDSTRUCTANONYMOUS({
uint32_t : 32;
});
});
});
} // namespace ucode

18
src/xenia/types.h
View File

@ -63,6 +63,24 @@ using std::tr1::shared_ptr;
#define XENOINLINE
#endif // MSVC
#if XE_COMPILER(MSVC)
#define XEPACKEDSTRUCT(name, value) \
__pragma(pack(push, 1)) struct name##_s value __pragma(pack(pop)); \
typedef struct name##_s name;
#define XEPACKEDSTRUCTANONYMOUS(value) \
__pragma(pack(push, 1)) struct value __pragma(pack(pop));
#define XEPACKEDUNION(name, value) \
__pragma(pack(push, 1)) union name##_s value __pragma(pack(pop)); \
typedef union name##_s name;
#elif XE_COMPILER(GNUC)
#define XEPACKEDSTRUCT(name, value) \
struct __attribute__((packed)) name
#define XEPACKEDSTRUCTANONYMOUS(value) \
struct __attribute__((packed))
#define XEPACKEDUNION(name, value) \
union __attribute__((packed)) name
#endif // MSVC
#if XE_COMPILER(MSVC)
// http://msdn.microsoft.com/en-us/library/83ythb65.aspx
#define XECACHEALIGN __declspec(align(XE_ALIGNMENT))