drm/radeon: group r6xx/r7xx newly sequential blit state

group state that is emitted sequentially into fewer packets.
This saves a number of dwords.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

1 files changed, 35 insertions, 203 deletions

diff --git a/drivers/gpu/drm/radeon/r600_blit_shaders.c b/drivers/gpu/drm/radeon/r600_blit_shaders.c
index 9a0c947b342c..762b81cb6bd3 100644
--- a/drivers/gpu/drm/radeon/r600_blit_shaders.c
+++ b/drivers/gpu/drm/radeon/r600_blit_shaders.c
@@ -109,22 +109,13 @@ const u32 r6xx_default_state[] =
 	0x00000351,
 	0x0000aa00, /* DB_ALPHA_TO_MASK */
 
-	0xc0036900,
+	0xc00f6900,
 	0x00000100,
 	0x00000800, /* VGT_MAX_VTX_INDX */
 	0x00000000, /* VGT_MIN_VTX_INDX */
 	0x00000000, /* VGT_INDX_OFFSET */
-
-	0xc0016900,
-	0x00000103,
 	0x00000000, /* VGT_MULTI_PRIM_IB_RESET_INDX */
-
-	0xc0016900,
-	0x00000104,
 	0x00000000, /* SX_ALPHA_TEST_CONTROL */
-
-	0xc0076900,
-	0x00000105,
 	0x00000000, /* CB_BLEND_RED */
 	0x00000000,
 	0x00000000,
@@ -132,16 +123,19 @@ const u32 r6xx_default_state[] =
 	0x00000000, /* CB_FOG_RED */
 	0x00000000,
 	0x00000000,
-
-	0xc0026900,
-	0x0000010c,
 	0x00000000, /* DB_STENCILREFMASK */
 	0x00000000, /* DB_STENCILREFMASK_BF */
-
-	0xc0016900,
-	0x0000010e,
 	0x00000000, /* SX_ALPHA_REF */
 
+	0xc0066900,
+	0x0000010f,
+	0x00000000, /* PA_CL_VPORT_XSCALE */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+
 	0xc0046900,
 	0x0000030c,
 	0x01000000, /* CB_CLRCMP_CNTL */
@@ -240,41 +234,20 @@ const u32 r6xx_default_state[] =
 	0x00000000,
 	0x3f800000,
 
-	0xc0016900,
+	0xc0026900,
 	0x00000292,
 	0x00000000, /* PA_SC_MPASS_PS_CNTL */
-
-	0xc0016900,
-	0x00000293,
 	0x00004010, /* PA_SC_MODE_CNTL */
 
-	0xc0066900,
-	0x0000010f,
-	0x00000000, /* PA_CL_VPORT_0_XSCALE */
-	0x00000000,
-	0x00000000,
-	0x00000000,
-	0x00000000,
-	0x00000000,
-
-	0xc0026900,
+	0xc0096900,
 	0x00000300,
 	0x00000000, /* PA_SC_LINE_CNTL */
 	0x00000000, /* PA_SC_AA_CONFIG */
-
-	0xc0016900,
-	0x00000302,
 	0x0000002d, /* PA_SU_VTX_CNTL */
-
-	0xc0046900,
-	0x00000303,
 	0x3f800000, /* PA_CL_GB_VERT_CLIP_ADJ */
 	0x3f800000,
 	0x3f800000,
 	0x3f800000,
-
-	0xc0026900,
-	0x00000307,
 	0x00000000, /* PA_SC_SAMPLE_LOCS_MCTX */
 	0x00000000,
 
@@ -282,28 +255,13 @@ const u32 r6xx_default_state[] =
 	0x00000312,
 	0xffffffff, /* PA_SC_AA_MASK */
 
-	0xc0016900,
+	0xc0066900,
 	0x0000037e,
 	0x00000000, /* PA_SU_POLY_OFFSET_DB_FMT_CNTL */
-
-	0xc0016900,
-	0x0000037f,
 	0x00000000, /* PA_SU_POLY_OFFSET_CLAMP */
-
-	0xc0016900,
-	0x00000380,
 	0x00000000, /* PA_SU_POLY_OFFSET_FRONT_SCALE */
-
-	0xc0016900,
-	0x00000381,
 	0x00000000, /* PA_SU_POLY_OFFSET_FRONT_OFFSET */
-
-	0xc0016900,
-	0x00000382,
 	0x00000000, /* PA_SU_POLY_OFFSET_BACK_SCALE */
-
-	0xc0016900,
-	0x00000383,
 	0x00000000, /* PA_SU_POLY_OFFSET_BACK_OFFSET */
 
 	0xc0046900,
@@ -330,29 +288,14 @@ const u32 r6xx_default_state[] =
 	0x00000000, /* VGT_INSTANCE_STEP_RATE_0 */
 	0x00000000, /* VGT_INSTANCE_STEP_RATE_1 */
 
-	0xc0026900,
+	0xc0116900,
 	0x00000280,
 	0x00000000, /* PA_SU_POINT_SIZE */
 	0x00000000, /* PA_SU_POINT_MINMAX */
-
-	0xc0016900,
-	0x00000282,
 	0x00000008, /* PA_SU_LINE_CNTL */
-
-	0xc0016900,
-	0x00000283,
 	0x00000000, /* PA_SC_LINE_STIPPLE */
-
-	0xc0016900,
-	0x00000284,
 	0x00000000, /* VGT_OUTPUT_PATH_CNTL */
-
-	0xc0016900,
-	0x00000285,
 	0x00000000, /* VGT_HOS_CNTL */
-
-	0xc00a6900,
-	0x00000286,
 	0x00000000, /* VGT_HOS_MAX_TESS_LEVEL */
 	0x00000000, /* VGT_HOS_MIN_TESS_LEVEL */
 	0x00000000, /* VGT_HOS_REUSE_DEPTH */
@@ -363,9 +306,6 @@ const u32 r6xx_default_state[] =
 	0x00000000, /* VGT_GROUP_VECT_1_CNTL */
 	0x00000000, /* VGT_GROUP_VECT_0_FMT_CNTL */
 	0x00000000, /* VGT_GROUP_VECT_1_FMT_CNTL */
-
-	0xc0016900,
-	0x00000290,
 	0x00000000, /* VGT_GS_MODE */
 
 	0xc0016900,
@@ -386,37 +326,19 @@ const u32 r6xx_default_state[] =
 	0x000002c8,
 	0x00000000, /* VGT_STRMOUT_BUFFER_EN */
 
-	0xc0016900,
+	0xc0076900,
 	0x00000202,
 	0x00cc0000, /* CB_COLOR_CONTROL */
-
-	0xc0016900,
-	0x00000203,
 	0x00000210, /* DB_SHADER_CNTL */
-
-	0xc0016900,
-	0x00000204,
 	0x00010000, /* PA_CL_CLIP_CNTL */
-
-	0xc0016900,
-	0x00000205,
 	0x00000244, /* PA_SU_SC_MODE_CNTL */
-
-	0xc0016900,
-	0x00000206,
 	0x00000100, /* PA_CL_VTE_CNTL */
-
-	0xc0026900,
-	0x00000207,
 	0x00000000, /* PA_CL_VS_OUT_CNTL */
 	0x00000000, /* PA_CL_NANINF_CNTL */
 
-	0xc0016900,
+	0xc0026900,
 	0x0000008e,
 	0x0000000f, /* CB_TARGET_MASK */
-
-	0xc0016900,
-	0x0000008f,
 	0x0000000f, /* CB_SHADER_MASK */
 
 	0xc0016900,
@@ -431,21 +353,12 @@ const u32 r6xx_default_state[] =
 	0x00000191,
 	0x00000b00, /* SPI_PS_INPUT_CNTL_0 */
 
-	0xc0016900,
+	0xc0056900,
 	0x000001b1,
 	0x00000000, /* SPI_VS_OUT_CONFIG */
-
-	0xc0016900,
-	0x000001b2,
 	0x00000000, /* SPI_THREAD_GROUPING */
-
-	0xc0026900,
-	0x000001b3,
 	0x00000001, /* SPI_PS_IN_CONTROL_0 */
 	0x00000000, /* SPI_PS_IN_CONTROL_1 */
-
-	0xc0016900,
-	0x000001b5,
 	0x00000000, /* SPI_INTERP_CONTROL_0 */
 
 	0xc0036e00, /* SET_SAMPLER */
@@ -524,36 +437,33 @@ const u32 r7xx_default_state[] =
 	0x00000351,
 	0x0000aa00, /* DB_ALPHA_TO_MASK */
 
-	0xc0036900,
+	0xc0096900,
 	0x00000100,
 	0x00000800, /* VGT_MAX_VTX_INDX */
 	0x00000000, /* VGT_MIN_VTX_INDX */
 	0x00000000, /* VGT_INDX_OFFSET */
-
-	0xc0016900,
-	0x00000103,
 	0x00000000, /* VGT_MULTI_PRIM_IB_RESET_INDX */
-
-	0xc0016900,
-	0x00000104,
 	0x00000000, /* SX_ALPHA_TEST_CONTROL */
-
-	0xc0046900,
-	0x00000105,
 	0x00000000, /* CB_BLEND_RED */
 	0x00000000,
 	0x00000000,
 	0x00000000,
 
-	0xc0026900,
+	0xc0036900,
 	0x0000010c,
 	0x00000000, /* DB_STENCILREFMASK */
 	0x00000000, /* DB_STENCILREFMASK_BF */
-
-	0xc0016900,
-	0x0000010e,
 	0x00000000, /* SX_ALPHA_REF */
 
+	0xc0066900,
+	0x0000010f,
+	0x00000000, /* PA_CL_VPORT_XSCALE */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+
 	0xc0046900,
 	0x0000030c, /* CB_CLRCMP_CNTL */
 	0x01000000,
@@ -645,41 +555,20 @@ const u32 r7xx_default_state[] =
 	0x00000000,
 	0x3f800000,
 
-	0xc0016900,
+	0xc0026900,
 	0x00000292,
 	0x00000000, /* PA_SC_MPASS_PS_CNTL */
-
-	0xc0016900,
-	0x00000293,
 	0x00514000, /* PA_SC_MODE_CNTL */
 
-	0xc0066900,
-	0x0000010f,
-	0x00000000, /* PA_CL_VPORT_0_XSCALE */
-	0x00000000,
-	0x00000000,
-	0x00000000,
-	0x00000000,
-	0x00000000,
-
-	0xc0026900,
+	0xc0096900,
 	0x00000300,
 	0x00000000, /* PA_SC_LINE_CNTL */
 	0x00000000, /* PA_SC_AA_CONFIG */
-
-	0xc0016900,
-	0x00000302,
 	0x0000002d, /* PA_SU_VTX_CNTL */
-
-	0xc0046900,
-	0x00000303,
 	0x3f800000, /* PA_CL_GB_VERT_CLIP_ADJ */
 	0x3f800000,
 	0x3f800000,
 	0x3f800000,
-
-	0xc0026900,
-	0x00000307,
 	0x00000000, /* PA_SC_SAMPLE_LOCS_MCTX */
 	0x00000000,
 
@@ -687,28 +576,13 @@ const u32 r7xx_default_state[] =
 	0x00000312,
 	0xffffffff, /* PA_SC_AA_MASK */
 
-	0xc0016900,
+	0xc0066900,
 	0x0000037e,
 	0x00000000, /* PA_SU_POLY_OFFSET_DB_FMT_CNTL */
-
-	0xc0016900,
-	0x0000037f,
 	0x00000000, /* PA_SU_POLY_OFFSET_CLAMP */
-
-	0xc0016900,
-	0x00000380,
 	0x00000000, /* PA_SU_POLY_OFFSET_FRONT_SCALE */
-
-	0xc0016900,
-	0x00000381,
 	0x00000000, /* PA_SU_POLY_OFFSET_FRONT_OFFSET */
-
-	0xc0016900,
-	0x00000382,
 	0x00000000, /* PA_SU_POLY_OFFSET_BACK_SCALE */
-
-	0xc0016900,
-	0x00000383,
 	0x00000000, /* PA_SU_POLY_OFFSET_BACK_OFFSET */
 
 	0xc0046900,
@@ -735,25 +609,13 @@ const u32 r7xx_default_state[] =
 	0x00000000, /* VGT_INSTANCE_STEP_RATE_0 */
 	0x00000000, /* VGT_INSTANCE_STEP_RATE_1 */
 
-	0xc0026900,
+	0xc0116900,
 	0x00000280,
 	0x00000000, /* PA_SU_POINT_SIZE */
 	0x00000000, /* PA_SU_POINT_MINMAX */
-
-	0xc0016900,
-	0x00000282,
 	0x00000008, /* PA_SU_LINE_CNTL */
-
-	0xc0016900,
-	0x00000283,
 	0x00000000, /* PA_SC_LINE_STIPPLE */
-
-	0xc0016900,
-	0x00000284,
 	0x00000000, /* VGT_OUTPUT_PATH_CNTL */
-
-	0xc00b6900,
-	0x00000285,
 	0x00000000, /* VGT_HOS_CNTL */
 	0x00000000, /* VGT_HOS_MAX_TESS_LEVEL */
 	0x00000000, /* VGT_HOS_MIN_TESS_LEVEL */
@@ -765,9 +627,6 @@ const u32 r7xx_default_state[] =
 	0x00000000, /* VGT_GROUP_VECT_1_CNTL */
 	0x00000000, /* VGT_GROUP_VECT_0_FMT_CNTL */
 	0x00000000, /* VGT_GROUP_VECT_1_FMT_CNTL */
-
-	0xc0016900,
-	0x00000290,
 	0x00000000, /* VGT_GS_MODE */
 
 	0xc0016900,
@@ -788,37 +647,19 @@ const u32 r7xx_default_state[] =
 	0x000002c8,
 	0x00000000, /* VGT_STRMOUT_BUFFER_EN */
 
-	0xc0016900,
+	0xc0076900,
 	0x00000202,
 	0x00cc0000, /* CB_COLOR_CONTROL */
-
-	0xc0016900,
-	0x00000203,
 	0x00000210, /* DB_SHADER_CNTL */
-
-	0xc0016900,
-	0x00000204,
 	0x00010000, /* PA_CL_CLIP_CNTL */
-
-	0xc0016900,
-	0x00000205,
 	0x00000244, /* PA_SU_SC_MODE_CNTL */
-
-	0xc0016900,
-	0x00000206,
 	0x00000100, /* PA_CL_VTE_CNTL */
-
-	0xc0026900,
-	0x00000207,
 	0x00000000, /* PA_CL_VS_OUT_CNTL */
 	0x00000000, /* PA_CL_NANINF_CNTL */
 
-	0xc0016900,
+	0xc0026900,
 	0x0000008e,
 	0x0000000f, /* CB_TARGET_MASK */
-
-	0xc0016900,
-	0x0000008f,
 	0x0000000f, /* CB_SHADER_MASK */
 
 	0xc0016900,
@@ -833,21 +674,12 @@ const u32 r7xx_default_state[] =
 	0x00000191,
 	0x00000b00, /* SPI_PS_INPUT_CNTL_0 */
 
-	0xc0016900,
+	0xc0056900,
 	0x000001b1,
 	0x00000000, /* SPI_VS_OUT_CONFIG */
-
-	0xc0016900,
-	0x000001b2,
 	0x00000001, /* SPI_THREAD_GROUPING */
-
-	0xc0026900,
-	0x000001b3,
 	0x00000001, /* SPI_PS_IN_CONTROL_0 */
 	0x00000000, /* SPI_PS_IN_CONTROL_1 */
-
-	0xc0016900,
-	0x000001b5,
 	0x00000000, /* SPI_INTERP_CONTROL_0 */
 
 	0xc0036e00, /* SET_SAMPLER */