radv: don't even attempt to prefetch on SI
authorGrazvydas Ignotas <notasas@gmail.com>
Sat, 10 Jun 2017 15:34:22 +0000 (18:34 +0300)
committerGrazvydas Ignotas <notasas@gmail.com>
Sun, 11 Jun 2017 11:28:40 +0000 (14:28 +0300)
Before bcae327469 this was emitting CP DMA packet even on SI, but
apparently hasn't caused too many problems. After that commit the
CP DMA code now always sets the CIK+ only bit for prefetch. Just
follow radeonsi there and don't try to prefetch at all.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101334
Fixes: bcae327469 "radv: realign cp dma code with radeonsi"
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
src/amd/vulkan/radv_cmd_buffer.c

index 1ac9de139f1f8f18be3fe23eeabad98681c32af1..b08f218549a3d918c066853ea2a6b7bf8eb4d487 100644 (file)
@@ -531,6 +531,14 @@ radv_emit_graphics_raster_state(struct radv_cmd_buffer *cmd_buffer,
                               raster->pa_su_sc_mode_cntl);
 }
 
+static inline void
+radv_emit_prefetch(struct radv_cmd_buffer *cmd_buffer, uint64_t va,
+                  unsigned size)
+{
+       if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK)
+               si_cp_dma_prefetch(cmd_buffer, va, size);
+}
+
 static void
 radv_emit_hw_vs(struct radv_cmd_buffer *cmd_buffer,
                struct radv_pipeline *pipeline,
@@ -542,7 +550,7 @@ radv_emit_hw_vs(struct radv_cmd_buffer *cmd_buffer,
        unsigned export_count;
 
        ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, shader->code_size);
+       radv_emit_prefetch(cmd_buffer, va, shader->code_size);
 
        export_count = MAX2(1, outinfo->param_exports);
        radeon_set_context_reg(cmd_buffer->cs, R_0286C4_SPI_VS_OUT_CONFIG,
@@ -591,7 +599,7 @@ radv_emit_hw_es(struct radv_cmd_buffer *cmd_buffer,
        uint64_t va = ws->buffer_get_va(shader->bo);
 
        ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, shader->code_size);
+       radv_emit_prefetch(cmd_buffer, va, shader->code_size);
 
        radeon_set_context_reg(cmd_buffer->cs, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
                               outinfo->esgs_itemsize / 4);
@@ -611,7 +619,7 @@ radv_emit_hw_ls(struct radv_cmd_buffer *cmd_buffer,
        uint32_t rsrc2 = shader->rsrc2;
 
        ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, shader->code_size);
+       radv_emit_prefetch(cmd_buffer, va, shader->code_size);
 
        radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B520_SPI_SHADER_PGM_LO_LS, 2);
        radeon_emit(cmd_buffer->cs, va >> 8);
@@ -635,7 +643,7 @@ radv_emit_hw_hs(struct radv_cmd_buffer *cmd_buffer,
        uint64_t va = ws->buffer_get_va(shader->bo);
 
        ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, shader->code_size);
+       radv_emit_prefetch(cmd_buffer, va, shader->code_size);
 
        radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B420_SPI_SHADER_PGM_LO_HS, 4);
        radeon_emit(cmd_buffer->cs, va >> 8);
@@ -769,7 +777,8 @@ radv_emit_geometry_shader(struct radv_cmd_buffer *cmd_buffer,
 
        va = ws->buffer_get_va(gs->bo);
        ws->cs_add_buffer(cmd_buffer->cs, gs->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, gs->code_size);
+       radv_emit_prefetch(cmd_buffer, va, gs->code_size);
+
        radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B220_SPI_SHADER_PGM_LO_GS, 4);
        radeon_emit(cmd_buffer->cs, va >> 8);
        radeon_emit(cmd_buffer->cs, va >> 40);
@@ -810,7 +819,7 @@ radv_emit_fragment_shader(struct radv_cmd_buffer *cmd_buffer,
 
        va = ws->buffer_get_va(ps->bo);
        ws->cs_add_buffer(cmd_buffer->cs, ps->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, ps->code_size);
+       radv_emit_prefetch(cmd_buffer, va, ps->code_size);
 
        radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B020_SPI_SHADER_PGM_LO_PS, 4);
        radeon_emit(cmd_buffer->cs, va >> 8);
@@ -2215,7 +2224,7 @@ radv_emit_compute_pipeline(struct radv_cmd_buffer *cmd_buffer)
        va = ws->buffer_get_va(compute_shader->bo);
 
        ws->cs_add_buffer(cmd_buffer->cs, compute_shader->bo, 8);
-       si_cp_dma_prefetch(cmd_buffer, va, compute_shader->code_size);
+       radv_emit_prefetch(cmd_buffer, va, compute_shader->code_size);
 
        MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
                                                           cmd_buffer->cs, 16);