From: Rhys Perry <pendingchaos02@gmail.com>
Date: Mon, 26 Sep 2022 16:18:48 +0000 (+0100)
Subject: aco/gfx11: increase gfx1100/gfx1101 physical vgprs
X-Git-Tag: upstream/22.3.5~561
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=50073d6135044b8a8ccc7ddaf35588594edeb3d1;p=platform%2Fupstream%2Fmesa.git

aco/gfx11: increase gfx1100/gfx1101 physical vgprs

https://reviews.llvm.org/D134522

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18825>
---

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 300e320..83a8537 100644
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -12563,7 +12563,7 @@ select_vs_prolog(Program* program, const struct aco_vs_prolog_key* key, ac_shade
    program->config->float_mode = program->blocks[0].fp_mode.val;
    /* addition on GFX6-8 requires a carry-out (we use VCC) */
    program->needs_vcc = program->gfx_level <= GFX8;
-   program->config->num_vgprs = get_vgpr_alloc(program, num_vgprs);
+   program->config->num_vgprs = std::min<uint16_t>(get_vgpr_alloc(program, num_vgprs), 256);
    program->config->num_sgprs = get_sgpr_alloc(program, num_sgprs);
 }
 
diff --git a/src/amd/compiler/aco_ir.cpp b/src/amd/compiler/aco_ir.cpp
index 14de439..84493d1 100644
--- a/src/amd/compiler/aco_ir.cpp
+++ b/src/amd/compiler/aco_ir.cpp
@@ -111,14 +111,20 @@ init_program(Program* program, Stage stage, const struct aco_shader_info* info,
 
    if (gfx_level >= GFX10) {
       program->dev.physical_sgprs = 5120; /* doesn't matter as long as it's at least 128 * 40 */
-      program->dev.physical_vgprs = program->wave_size == 32 ? 1024 : 512;
       program->dev.sgpr_alloc_granule = 128;
       program->dev.sgpr_limit =
          108; /* includes VCC, which can be treated as s[106-107] on GFX10+ */
-      if (gfx_level == GFX10_3)
-         program->dev.vgpr_alloc_granule = program->wave_size == 32 ? 16 : 8;
-      else
-         program->dev.vgpr_alloc_granule = program->wave_size == 32 ? 8 : 4;
+
+      if (family == CHIP_GFX1100 || family == CHIP_GFX1101) {
+         program->dev.physical_vgprs = program->wave_size == 32 ? 1536 : 768;
+         program->dev.vgpr_alloc_granule = program->wave_size == 32 ? 24 : 12;
+      } else {
+         program->dev.physical_vgprs = program->wave_size == 32 ? 1024 : 512;
+         if (gfx_level >= GFX10_3)
+            program->dev.vgpr_alloc_granule = program->wave_size == 32 ? 16 : 8;
+         else
+            program->dev.vgpr_alloc_granule = program->wave_size == 32 ? 8 : 4;
+      }
    } else if (program->gfx_level >= GFX8) {
       program->dev.physical_sgprs = 800;
       program->dev.sgpr_alloc_granule = 16;
diff --git a/src/amd/compiler/aco_ir.h b/src/amd/compiler/aco_ir.h
index b2c9db4..9a10c67 100644
--- a/src/amd/compiler/aco_ir.h
+++ b/src/amd/compiler/aco_ir.h
@@ -2134,7 +2134,7 @@ struct DeviceInfo {
    uint16_t vgpr_limit;
    uint16_t sgpr_limit;
    uint16_t sgpr_alloc_granule;
-   uint16_t vgpr_alloc_granule; /* must be power of two */
+   uint16_t vgpr_alloc_granule;
    unsigned max_wave64_per_simd;
    unsigned simd_per_cu;
    bool has_fast_fma32 = false;
diff --git a/src/amd/compiler/aco_live_var_analysis.cpp b/src/amd/compiler/aco_live_var_analysis.cpp
index 80bbff4..0ecc957 100644
--- a/src/amd/compiler/aco_live_var_analysis.cpp
+++ b/src/amd/compiler/aco_live_var_analysis.cpp
@@ -348,7 +348,7 @@ get_vgpr_alloc(Program* program, uint16_t addressable_vgprs)
 {
    assert(addressable_vgprs <= program->dev.vgpr_limit);
    uint16_t granule = program->dev.vgpr_alloc_granule;
-   return align(std::max(addressable_vgprs, granule), granule);
+   return ALIGN_NPOT(std::max(addressable_vgprs, granule), granule);
 }
 
 unsigned
@@ -370,7 +370,8 @@ get_addr_sgpr_from_waves(Program* program, uint16_t waves)
 uint16_t
 get_addr_vgpr_from_waves(Program* program, uint16_t waves)
 {
-   uint16_t vgprs = program->dev.physical_vgprs / waves & ~(program->dev.vgpr_alloc_granule - 1);
+   uint16_t vgprs = program->dev.physical_vgprs / waves;
+   vgprs = vgprs / program->dev.vgpr_alloc_granule * program->dev.vgpr_alloc_granule;
    vgprs -= program->config->num_shared_vgprs / 2;
    return std::min(vgprs, program->dev.vgpr_limit);
 }
diff --git a/src/amd/compiler/aco_register_allocation.cpp b/src/amd/compiler/aco_register_allocation.cpp
index 0a3401e..eef4ff0 100644
--- a/src/amd/compiler/aco_register_allocation.cpp
+++ b/src/amd/compiler/aco_register_allocation.cpp
@@ -3149,7 +3149,8 @@ register_allocation(Program* program, std::vector<IDSet>& live_out_per_block, ra
    } /* end for BB */
 
    /* num_gpr = rnd_up(max_used_gpr + 1) */
-   program->config->num_vgprs = get_vgpr_alloc(program, ctx.max_used_vgpr + 1);
+   program->config->num_vgprs =
+      std::min<uint16_t>(get_vgpr_alloc(program, ctx.max_used_vgpr + 1), 256);
    program->config->num_sgprs = get_sgpr_alloc(program, ctx.max_used_sgpr + 1);
 
    program->progress = CompilationProgress::after_ra;