radeonsi: do EarlyCSEMemSSA LLVM pass

so that LLVM IR looks like CSE has been run on it. It's also recommended
by the instruction combining pass.

This also fixes:
- GL45-CTS.arrays_of_arrays_gl.InteractionFunctionCalls2 (crash)
- piglit/spec/arb_shader_ballot/execution/fs-readFirstInvocation-uint-loop (fail)

The code size decrease is positive, the register usage isn't. There is
a decrease in VGPR spilling for Tomb Raider, but increase in DiRT Showdown
and GRID Autosport.

EarlyCSEMemSSA has a -0.01% change in code size compared EarlyCSE.

SGPRS: 1935420 -> 1938076 (0.14 %)
VGPRS: 1645504 -> 1645988 (0.03 %)
Spilled SGPRs: 2493 -> 2651 (6.34 %)
Spilled VGPRs: 107 -> 115 (7.48 %)
Private memory VGPRs: 1332 -> 1332 (0.00 %)
Scratch size: 1512 -> 1516 (0.26 %) dwords per thread
Code Size: 61981592 -> 61890012 (-0.15 %) bytes
Max Waves: 371847 -> 371798 (-0.01 %)

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
index 1f8e91343f9d5ba5b7f27e971c686239d207f121..802e2b94e9ea92642b4d76f4a6d939a081ca3335 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
@@ -1451,6 +1451,8 @@ void si_llvm_optimize_module(struct si_shader_context *ctx)
        LLVMAddLICMPass(gallivm->passmgr);
        LLVMAddAggressiveDCEPass(gallivm->passmgr);
        LLVMAddCFGSimplificationPass(gallivm->passmgr);
+       /* This is recommended by the instruction combining pass. */
+       LLVMAddEarlyCSEMemSSAPass(gallivm->passmgr);
        LLVMAddInstructionCombiningPass(gallivm->passmgr);
 
        /* Run the pass */