flywithu/ntel-gpu-tools
1
0
Fork 0
mirror of https://github.com/tiagovignatti/intel-gpu-tools.git synced 2025-06-11 09:56:22 +00:00
Code Issues Releases Wiki Activity

assembler: Implement register-indirect addressing mode in brw_set_src1()

Browse Source 
The assembler allows people to do that and that's something available
since Crestline.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
This commit is contained in:
Damien Lespiau 2013-01-28 15:27:59 +00:00
parent 8eb30d9493
commit d3a2a678d5
1 changed files with 27 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
assembler/brw_eu_emit.c
View File

@ -351,6 +351,9 @@ void brw_set_src1(struct brw_compile *p,
struct brw_instruction *insn, struct brw_instruction *insn,
struct brw_reg reg) struct brw_reg reg)
{ {
struct brw_context *brw = p->brw;
struct intel_context *intel = &brw->intel;
assert(reg.file != BRW_MESSAGE_REGISTER_FILE); assert(reg.file != BRW_MESSAGE_REGISTER_FILE);
if (reg.file != BRW_ARCHITECTURE_REGISTER_FILE) if (reg.file != BRW_ARCHITECTURE_REGISTER_FILE)
@ -364,6 +367,7 @@ void brw_set_src1(struct brw_compile *p,
insn->bits1.da1.src1_reg_type = reg.type; insn->bits1.da1.src1_reg_type = reg.type;
insn->bits3.da1.src1_abs = reg.abs; insn->bits3.da1.src1_abs = reg.abs;
insn->bits3.da1.src1_negate = reg.negate; insn->bits3.da1.src1_negate = reg.negate;
insn->bits3.da1.src1_address_mode = reg.address_mode;
/* Only src1 can be immediate in two-argument instructions. /* Only src1 can be immediate in two-argument instructions.
*/ */
@ -373,33 +377,44 @@ void brw_set_src1(struct brw_compile *p,
insn->bits3.ud = reg.dw1.ud; insn->bits3.ud = reg.dw1.ud;
} }
else { else {
/* This is a hardware restriction, which may or may not be lifted /* It's only BRW that does not support register-indirect addressing on
* in the future: * src1 */
*/ assert (intel->gen >= 4 || reg.address_mode == BRW_ADDRESS_DIRECT);
assert (reg.address_mode == BRW_ADDRESS_DIRECT);
/* assert (reg.file == BRW_GENERAL_REGISTER_FILE); */
if (insn->header.access_mode == BRW_ALIGN_1) { if (reg.address_mode == BRW_ADDRESS_DIRECT) {
insn->bits3.da1.src1_subreg_nr = reg.subnr; if (insn->header.access_mode == BRW_ALIGN_1) {
insn->bits3.da1.src1_reg_nr = reg.nr; insn->bits3.da1.src1_subreg_nr = reg.subnr;
insn->bits3.da1.src1_reg_nr = reg.nr;
}
else {
insn->bits3.da16.src1_subreg_nr = reg.subnr / 16;
insn->bits3.da16.src1_reg_nr = reg.nr;
}
} }
else { else {
insn->bits3.da16.src1_subreg_nr = reg.subnr / 16; insn->bits3.ia1.src1_subreg_nr = reg.subnr;
insn->bits3.da16.src1_reg_nr = reg.nr;
if (insn->header.access_mode == BRW_ALIGN_1)
insn->bits3.ia1.src1_indirect_offset = reg.dw1.bits.indirect_offset;
else
insn->bits3.ia16.src1_indirect_offset = reg.dw1.bits.indirect_offset / 16;
} }
if (insn->header.access_mode == BRW_ALIGN_1) { if (insn->header.access_mode == BRW_ALIGN_1) {
/* FIXME: While this is correct, if the assembler uses that code path
* the opcode generated are different and thus needs a validation
* pass.
if (reg.width == BRW_WIDTH_1 && if (reg.width == BRW_WIDTH_1 &&
insn->header.execution_size == BRW_EXECUTE_1) { insn->header.execution_size == BRW_EXECUTE_1) {
insn->bits3.da1.src1_horiz_stride = BRW_HORIZONTAL_STRIDE_0; insn->bits3.da1.src1_horiz_stride = BRW_HORIZONTAL_STRIDE_0;
insn->bits3.da1.src1_width = BRW_WIDTH_1; insn->bits3.da1.src1_width = BRW_WIDTH_1;
insn->bits3.da1.src1_vert_stride = BRW_VERTICAL_STRIDE_0; insn->bits3.da1.src1_vert_stride = BRW_VERTICAL_STRIDE_0;
} }
else { else { */
insn->bits3.da1.src1_horiz_stride = reg.hstride; insn->bits3.da1.src1_horiz_stride = reg.hstride;
insn->bits3.da1.src1_width = reg.width; insn->bits3.da1.src1_width = reg.width;
insn->bits3.da1.src1_vert_stride = reg.vstride; insn->bits3.da1.src1_vert_stride = reg.vstride;
} /* } */
} }
else { else {
insn->bits3.da16.src1_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X); insn->bits3.da16.src1_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X);