radeonsi/gfx10: disable LATE_ALLOC_GS on Navi14

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 64ce81f989cf6f3bc6b1536d88e29f99a1d6b2ed..66f0ba8f2019b2d5f12175894fdb9ed69594eeac 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -5545,6 +5545,7 @@ static void si_init_config(struct si_context *sctx)
                        late_alloc_limit = (num_cu_per_sh - 2) * 4;
                }
 
+               unsigned late_alloc_limit_gs = late_alloc_limit;
                unsigned cu_mask_vs = 0xffff;
                unsigned cu_mask_gs = 0xffff;
 
@@ -5558,6 +5559,12 @@ static void si_init_config(struct si_context *sctx)
                        }
                }
 
+               /* Don't use late alloc for NGG on Navi14 due to a hw bug. */
+               if (sctx->family == CHIP_NAVI14) {
+                       late_alloc_limit_gs = 0;
+                       cu_mask_gs = 0xffff;
+               }
+
                /* VS can't execute on one CU if the limit is > 2. */
                si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS,
                        S_00B118_CU_EN(cu_mask_vs) |
@@ -5571,7 +5578,7 @@ static void si_init_config(struct si_context *sctx)
                if (sctx->chip_class >= GFX10) {
                        si_pm4_set_reg(pm4, R_00B204_SPI_SHADER_PGM_RSRC4_GS,
                                       S_00B204_CU_EN(0xffff) |
-                                      S_00B204_SPI_SHADER_LATE_ALLOC_GS_GFX10(late_alloc_limit));
+                                      S_00B204_SPI_SHADER_LATE_ALLOC_GS_GFX10(late_alloc_limit_gs));
                }
 
                si_pm4_set_reg(pm4, R_00B01C_SPI_SHADER_PGM_RSRC3_PS,