r300: Set correct VAP_CNTL per vertex program.

diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c
index 1b405889c3c7016b0674bf6e861fb6fbd3d31b26..baa741816cbf2e9ff20f38fb094d01cf8f319148 100644 (file)
--- a/src/mesa/drivers/dri/r300/r300_ioctl.c
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.c
@@ -316,6 +316,14 @@ static void r300EmitClearState(GLcontext * ctx)
        e32(FP_SELA(0, NO, W, FP_TMP(0), 0, 0));
 
        if (has_tcl) {
+               R300_STATECHANGE(rmesa, vap_cntl);
+               reg_start(R300_VAP_CNTL, 0);
+
+               e32((10 << R300_VAP_CNTL__PVS_NUM_SLOTS__SHIFT) |
+                   (6 << R300_VAP_CNTL__PVS_NUM_CNTRLS__SHIFT) |
+                   (4 << R300_VAP_CNTL__PVS_NUM_FPUS__SHIFT) |
+                   (12 << R300_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT));
+
                R300_STATECHANGE(r300, pvs);
                reg_start(R300_VAP_PVS_CNTL_1, 2);
 

diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h
index 2200cec6abe6f133a7be2662b4b6cfaf7fb401a9..d2a81750fb4c59e84c4023f1a611ef56eccd1afd 100644 (file)
--- a/src/mesa/drivers/dri/r300/r300_reg.h
+++ b/src/mesa/drivers/dri/r300/r300_reg.h
@@ -67,9 +67,15 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 /*
  * Vertex Array Processing (VAP) Control
- * Stolen from r200 code from Christoph Brill (It's a guess!)
  */
 #define R300_VAP_CNTL  0x2080
+#      define  R300_VAP_CNTL__PVS_NUM_SLOTS__SHIFT             0
+#      define  R300_VAP_CNTL__PVS_NUM_CNTRLS__SHIFT            4
+#      define  R300_VAP_CNTL__PVS_NUM_FPUS__SHIFT              8
+#      define  R300_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT            18
+#      define  R500_VAP_CNTL__VAP_NO_RENDER                    (1<<17)
+#      define  R300_VAP_CNTL__DX_CLIP_SPACE_DEF                (1<<22)
+#      define  R500_VAP_CNTL__TCL_STATE_OPTIMIZATION           (1<<23)
 
 /* This register is written directly and also starts data section
  * in many 3d CP_PACKET3's

diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index e11b5afc30c468f1afe10c393ed294818b7099d3..ae5d8a80623236dc07da24c2bf03f3830b09f8b3 100644 (file)
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1648,10 +1648,51 @@ static inline void r300SetupVertexProgramFragment(r300ContextPtr r300, int dest,
        }
 }
 
+/* FIXME: move near the MIN2 define. */
+#define MIN3(a, b, c)  ((a) < (b) ? MIN2(a, c) : MIN2(b, c))
+
+/* FIXME: need to add a structure for per-card/chipset values; they are
+ * currently hard-coded. */
+static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
+                       GLuint output_count, GLuint temp_count)
+{
+       int cmd_reserved = 0;
+       int cmd_written = 0;
+       drm_radeon_cmd_header_t *cmd = NULL;
+
+       int vtx_mem_size = 72;  /* FIXME: R3XX vs R5XX */
+
+       /* Flush PVS engine before changing PVS_NUM_SLOTS, PVS_NUM_CNTRLS.
+        * See r500 docs 6.5.2 */
+       reg_start(R300_VAP_PVS_WAITIDLE, 0);
+       e32(0x00000000);
+
+       /* avoid division by zero */
+       if (input_count == 0)
+               input_count = 1;
+       if (output_count == 0)
+               output_count = 1;
+       if (temp_count == 0)
+               temp_count = 1;
+
+       int pvs_num_slots =
+           MIN3(10, vtx_mem_size / input_count, vtx_mem_size / output_count);
+       int pvs_num_cntrls = MIN2(6, vtx_mem_size / temp_count);
+
+       R300_STATECHANGE(rmesa, vap_cntl);
+       rmesa->hw.vap_cntl.cmd[1] =
+           (pvs_num_slots << R300_VAP_CNTL__PVS_NUM_SLOTS__SHIFT) |
+           (pvs_num_cntrls << R300_VAP_CNTL__PVS_NUM_CNTRLS__SHIFT) |
+           (4 << R300_VAP_CNTL__PVS_NUM_FPUS__SHIFT) |
+           (12 << R300_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT) |
+           R500_VAP_CNTL__TCL_STATE_OPTIMIZATION;
+}
+
 static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)
 {
        struct r300_vertex_shader_state *prog = &(rmesa->state.vertex_shader);
        GLuint o_reg = 0;
+       GLuint i_reg = 0;
        int i;
        int inst_count = 0;
        int param_count = 0;
@@ -1664,6 +1705,7 @@ static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)
                        prog->program.body.i[program_end + 2] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
                        prog->program.body.i[program_end + 3] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
                        program_end += 4;
+                       i_reg++;
                }
        }
 
@@ -1673,6 +1715,8 @@ static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)
                                       &(prog->program));
        inst_count = (prog->program.length / 4) - 1;
 
+       r300VapCntl(rmesa, i_reg, o_reg, 0);
+
        R300_STATECHANGE(rmesa, pvs);
        rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] =
            (0 << R300_PVS_CNTL_1_PROGRAM_START_SHIFT) |
@@ -1686,6 +1730,15 @@ static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)
            (inst_count << R300_PVS_CNTL_3_PROGRAM_UNKNOWN2_SHIFT);
 }
 
+static int r300BitCount(int x)
+{
+       x = ((x & 0xaaaaaaaaU) >> 1) + (x & 0x55555555U);
+       x = ((x & 0xccccccccU) >> 2) + (x & 0x33333333U);
+       x = (x >> 16) + (x & 0xffff);
+       x = ((x & 0xf0f0) >> 4) + (x & 0x0f0f);
+       return (x >> 8) + (x & 0x00ff);
+}
+
 static void r300SetupRealVertexProgram(r300ContextPtr rmesa)
 {
        GLcontext *ctx = rmesa->radeon.glCtx;
@@ -1707,6 +1760,10 @@ static void r300SetupRealVertexProgram(r300ContextPtr rmesa)
        r300SetupVertexProgramFragment(rmesa, R300_PVS_UPLOAD_PROGRAM, &(prog->program));
        inst_count = (prog->program.length / 4) - 1;
 
+       r300VapCntl(rmesa, r300BitCount(prog->key.InputsRead),
+                   r300BitCount(prog->key.OutputsWritten),
+                   prog->num_temporaries);
+       
        R300_STATECHANGE(rmesa, pvs);
        rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] =
          (0 << R300_PVS_CNTL_1_PROGRAM_START_SHIFT) |
@@ -1740,13 +1797,6 @@ static void r300SetupVertexProgram(r300ContextPtr rmesa)
                r300SetupDefaultVertexProgram(rmesa);
        }
 
-
-       /* FIXME: This is done for vertex shader fragments, but also needs to be
-        * done for vap_pvs, so I leave it as a reminder. */
-#if 0
-       reg_start(R300_VAP_PVS_WAITIDLE, 0);
-       e32(0x00000000);
-#endif
 }
 
 /**
@@ -1848,11 +1898,6 @@ static void r300ResetHwState(r300ContextPtr r300)
        r300AlphaFunc(ctx, ctx->Color.AlphaFunc, ctx->Color.AlphaRef);
        r300Enable(ctx, GL_ALPHA_TEST, ctx->Color.AlphaEnabled);
 
-       if (!has_tcl)
-               r300->hw.vap_cntl.cmd[1] = 0x0014045a;
-       else
-               r300->hw.vap_cntl.cmd[1] = 0x0030045A;  //0x0030065a /* Dangerous */
-
        r300->hw.vte.cmd[1] = R300_VPORT_X_SCALE_ENA
            | R300_VPORT_X_OFFSET_ENA
            | R300_VPORT_Y_SCALE_ENA