Mesa (master): aco: ensure readfirstlane subdword operands are always dword aligned

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 1b3be07b5faf867f698668080b060a270c5f795e
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=1b3be07b5faf867f698668080b060a270c5f795e

Author: Daniel Schürmann <daniel at schuermann.dev>
Date:   Thu Jul 23 09:58:11 2020 +0200

aco: ensure readfirstlane subdword operands are always dword aligned

Cc: 20.1 <mesa-stable at lists.freedesktop.org>
Reviewed-by: Rhys Perry <pendingchaos02 at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6024>

---

 src/amd/compiler/aco_register_allocation.cpp | 3 +++
 src/amd/compiler/aco_validate.cpp            | 2 ++
 2 files changed, 5 insertions(+)

diff --git a/src/amd/compiler/aco_register_allocation.cpp b/src/amd/compiler/aco_register_allocation.cpp
index 2fd8d3d66d5..862b00d859b 100644
--- a/src/amd/compiler/aco_register_allocation.cpp
+++ b/src/amd/compiler/aco_register_allocation.cpp
@@ -318,6 +318,9 @@ void print_regs(ra_ctx& ctx, bool vgprs, RegisterFile& reg_file)
 
 unsigned get_subdword_operand_stride(chip_class chip, const aco_ptr<Instruction>& instr, unsigned idx, RegClass rc)
 {
+   /* v_readfirstlane_b32 cannot use SDWA */
+   if (instr->opcode == aco_opcode::p_as_uniform)
+      return 4;
    if (instr->format == Format::PSEUDO && chip >= GFX8)
       return rc.bytes() % 2 == 0 ? 2 : 1;
 
diff --git a/src/amd/compiler/aco_validate.cpp b/src/amd/compiler/aco_validate.cpp
index e532cdf241f..97967aac9c1 100644
--- a/src/amd/compiler/aco_validate.cpp
+++ b/src/amd/compiler/aco_validate.cpp
@@ -483,6 +483,8 @@ bool validate_subdword_operand(chip_class chip, const aco_ptr<Instruction>& inst
    Operand op = instr->operands[index];
    unsigned byte = op.physReg().byte();
 
+   if (instr->opcode == aco_opcode::p_as_uniform)
+      return byte == 0;
    if (instr->format == Format::PSEUDO && chip >= GFX8)
       return true;
    if (instr->isSDWA() && (static_cast<SDWA_instruction *>(instr.get())->sel[index] & sdwa_asuint) == (sdwa_isra | op.bytes()))