AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aco: fix NSA following writelane

Browse Source 
No fossil-db changes on GFX10.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Fixes: c353895c92 ("aco: use non-sequential addressing")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9187>
This commit is contained in:
Rhys Perry
2021-03-10 17:42:52 +00:00
committed by Marge Bot
parent 298d400e5c
commit 502a073552
1 changed files with 14 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_insert_NOPs.cpp
+14
View File
@@ -151,6 +151,7 @@ struct NOP_ctx_gfx10 {
bool has_DS = false;
bool has_branch_after_DS = false;
bool has_NSA_MIMG = false;
bool has_writelane = false;
std::bitset<128> sgprs_read_by_VMEM;
std::bitset<128> sgprs_read_by_SMEM;
@@ -162,6 +163,7 @@ struct NOP_ctx_gfx10 {
has_DS |= other.has_DS;
has_branch_after_DS |= other.has_branch_after_DS;
has_NSA_MIMG |= other.has_NSA_MIMG;
has_writelane |= other.has_writelane;
sgprs_read_by_VMEM |= other.sgprs_read_by_VMEM;
sgprs_read_by_SMEM |= other.sgprs_read_by_SMEM;
}
@@ -176,6 +178,7 @@ struct NOP_ctx_gfx10 {
has_DS == other.has_DS &&
has_branch_after_DS == other.has_branch_after_DS &&
has_NSA_MIMG == other.has_NSA_MIMG &&
has_writelane == other.has_writelane &&
sgprs_read_by_VMEM == other.sgprs_read_by_VMEM &&
sgprs_read_by_SMEM == other.sgprs_read_by_SMEM;
}
@@ -756,6 +759,17 @@ void handle_instruction_gfx10(Program *program, Block *cur_block, NOP_ctx_gfx10
Builder(program, &new_instructions).sopp(aco_opcode::s_nop, -1, 0);
}
}
/* waNsaCannotFollowWritelane
* Handles NSA MIMG immediately following a v_writelane_b32.
*/
if (instr->opcode == aco_opcode::v_writelane_b32_e64) {
ctx.has_writelane = true;
} else if (ctx.has_writelane) {
ctx.has_writelane = false;
if (instr->isMIMG() && get_mimg_nsa_dwords(instr.get()) > 0)
Builder(program, &new_instructions).sopp(aco_opcode::s_nop, -1, 0);
}
}
template <typename Ctx>