Merge remote-tracking branch 'mesa-public/master' into vulkan
[mesa.git] / src / mesa / drivers / dri / i965 / brw_eu.c
index 8c7235a2b17f1772b62eeeccad1b6aa5ccf27122..1f4a3516fa2319273e2c5e14082df20c1b957dc2 100644 (file)
@@ -1,6 +1,6 @@
 /*
  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ Intel funded Tungsten Graphics to
  develop this 3D driver.
 
  Permission is hereby granted, free of charge, to any person obtaining
@@ -26,7 +26,7 @@
  **********************************************************************/
  /*
   * Authors:
-  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Keith Whitwell <keithw@vmware.com>
   */
 
 
 #include "brw_defines.h"
 #include "brw_eu.h"
 
-#include "glsl/ralloc.h"
+#include "util/ralloc.h"
+
+/**
+ * Converts a BRW_REGISTER_TYPE_* enum to a short string (F, UD, and so on).
+ *
+ * This is different than reg_encoding from brw_disasm.c in that it operates
+ * on the abstract enum values, rather than the generation-specific encoding.
+ */
+const char *
+brw_reg_type_letters(unsigned type)
+{
+   const char *names[] = {
+      [BRW_REGISTER_TYPE_UD] = "UD",
+      [BRW_REGISTER_TYPE_D]  = "D",
+      [BRW_REGISTER_TYPE_UW] = "UW",
+      [BRW_REGISTER_TYPE_W]  = "W",
+      [BRW_REGISTER_TYPE_F]  = "F",
+      [BRW_REGISTER_TYPE_UB] = "UB",
+      [BRW_REGISTER_TYPE_B]  = "B",
+      [BRW_REGISTER_TYPE_UV] = "UV",
+      [BRW_REGISTER_TYPE_V]  = "V",
+      [BRW_REGISTER_TYPE_VF] = "VF",
+      [BRW_REGISTER_TYPE_DF] = "DF",
+      [BRW_REGISTER_TYPE_HF] = "HF",
+      [BRW_REGISTER_TYPE_UQ] = "UQ",
+      [BRW_REGISTER_TYPE_Q]  = "Q",
+   };
+   assert(type <= BRW_REGISTER_TYPE_Q);
+   return names[type];
+}
+
+/* Returns a conditional modifier that negates the condition. */
+enum brw_conditional_mod
+brw_negate_cmod(uint32_t cmod)
+{
+   switch (cmod) {
+   case BRW_CONDITIONAL_Z:
+      return BRW_CONDITIONAL_NZ;
+   case BRW_CONDITIONAL_NZ:
+      return BRW_CONDITIONAL_Z;
+   case BRW_CONDITIONAL_G:
+      return BRW_CONDITIONAL_LE;
+   case BRW_CONDITIONAL_GE:
+      return BRW_CONDITIONAL_L;
+   case BRW_CONDITIONAL_L:
+      return BRW_CONDITIONAL_GE;
+   case BRW_CONDITIONAL_LE:
+      return BRW_CONDITIONAL_G;
+   default:
+      return ~0;
+   }
+}
 
 /* Returns the corresponding conditional mod for swapping src0 and
  * src1 in e.g. CMP.
  */
-uint32_t
+enum brw_conditional_mod
 brw_swap_cmod(uint32_t cmod)
 {
    switch (cmod) {
@@ -55,64 +106,47 @@ brw_swap_cmod(uint32_t cmod)
    case BRW_CONDITIONAL_LE:
       return BRW_CONDITIONAL_GE;
    default:
-      return ~0;
+      return BRW_CONDITIONAL_NONE;
    }
 }
 
-
-/* How does predicate control work when execution_size != 8?  Do I
- * need to test/set for 0xffff when execution_size is 16?
- */
-void brw_set_predicate_control_flag_value( struct brw_compile *p, unsigned value )
+void
+brw_set_default_exec_size(struct brw_codegen *p, unsigned value)
 {
-   p->current->header.predicate_control = BRW_PREDICATE_NONE;
-
-   if (value != 0xff) {
-      if (value != p->flag_value) {
-        brw_push_insn_state(p);
-        brw_MOV(p, brw_flag_reg(0, 0), brw_imm_uw(value));
-        p->flag_value = value;
-        brw_pop_insn_state(p);
-      }
-
-      p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
-   }
+   brw_inst_set_exec_size(p->devinfo, p->current, value);
 }
 
-void brw_set_predicate_control( struct brw_compile *p, unsigned pc )
+void brw_set_default_predicate_control( struct brw_codegen *p, unsigned pc )
 {
-   p->current->header.predicate_control = pc;
+   brw_inst_set_pred_control(p->devinfo, p->current, pc);
 }
 
-void brw_set_predicate_inverse(struct brw_compile *p, bool predicate_inverse)
+void brw_set_default_predicate_inverse(struct brw_codegen *p, bool predicate_inverse)
 {
-   p->current->header.predicate_inverse = predicate_inverse;
+   brw_inst_set_pred_inv(p->devinfo, p->current, predicate_inverse);
 }
 
-void brw_set_conditionalmod( struct brw_compile *p, unsigned conditional )
+void brw_set_default_flag_reg(struct brw_codegen *p, int reg, int subreg)
 {
-   p->current->header.destreg__conditionalmod = conditional;
-}
+   if (p->devinfo->gen >= 7)
+      brw_inst_set_flag_reg_nr(p->devinfo, p->current, reg);
 
-void brw_set_flag_reg(struct brw_compile *p, int reg, int subreg)
-{
-   p->current->bits2.da1.flag_reg_nr = reg;
-   p->current->bits2.da1.flag_subreg_nr = subreg;
+   brw_inst_set_flag_subreg_nr(p->devinfo, p->current, subreg);
 }
 
-void brw_set_access_mode( struct brw_compile *p, unsigned access_mode )
+void brw_set_default_access_mode( struct brw_codegen *p, unsigned access_mode )
 {
-   p->current->header.access_mode = access_mode;
+   brw_inst_set_access_mode(p->devinfo, p->current, access_mode);
 }
 
 void
-brw_set_compression_control(struct brw_compile *p,
+brw_set_default_compression_control(struct brw_codegen *p,
                            enum brw_compression compression_control)
 {
    p->compressed = (compression_control == BRW_COMPRESSION_COMPRESSED);
 
-   if (p->brw->gen >= 6) {
-      /* Since we don't use the 32-wide support in gen6, we translate
+   if (p->devinfo->gen >= 6) {
+      /* Since we don't use the SIMD32 support in gen6, we translate
        * the pre-gen6 compression control here.
        */
       switch (compression_control) {
@@ -120,53 +154,51 @@ brw_set_compression_control(struct brw_compile *p,
         /* This is the "use the first set of bits of dmask/vmask/arf
          * according to execsize" option.
          */
-        p->current->header.compression_control = GEN6_COMPRESSION_1Q;
+         brw_inst_set_qtr_control(p->devinfo, p->current, GEN6_COMPRESSION_1Q);
         break;
       case BRW_COMPRESSION_2NDHALF:
-        /* For 8-wide, this is "use the second set of 8 bits." */
-        p->current->header.compression_control = GEN6_COMPRESSION_2Q;
+        /* For SIMD8, this is "use the second set of 8 bits." */
+         brw_inst_set_qtr_control(p->devinfo, p->current, GEN6_COMPRESSION_2Q);
         break;
       case BRW_COMPRESSION_COMPRESSED:
-        /* For 16-wide instruction compression, use the first set of 16 bits
-         * since we don't do 32-wide dispatch.
+        /* For SIMD16 instruction compression, use the first set of 16 bits
+         * since we don't do SIMD32 dispatch.
          */
-        p->current->header.compression_control = GEN6_COMPRESSION_1H;
+         brw_inst_set_qtr_control(p->devinfo, p->current, GEN6_COMPRESSION_1H);
         break;
       default:
-        assert(!"not reached");
-        p->current->header.compression_control = GEN6_COMPRESSION_1H;
-        break;
+         unreachable("not reached");
       }
    } else {
-      p->current->header.compression_control = compression_control;
+      brw_inst_set_qtr_control(p->devinfo, p->current, compression_control);
    }
 }
 
-void brw_set_mask_control( struct brw_compile *p, unsigned value )
+void brw_set_default_mask_control( struct brw_codegen *p, unsigned value )
 {
-   p->current->header.mask_control = value;
+   brw_inst_set_mask_control(p->devinfo, p->current, value);
 }
 
-void brw_set_saturate( struct brw_compile *p, bool enable )
+void brw_set_default_saturate( struct brw_codegen *p, bool enable )
 {
-   p->current->header.saturate = enable;
+   brw_inst_set_saturate(p->devinfo, p->current, enable);
 }
 
-void brw_set_acc_write_control(struct brw_compile *p, unsigned value)
+void brw_set_default_acc_write_control(struct brw_codegen *p, unsigned value)
 {
-   if (p->brw->gen >= 6)
-      p->current->header.acc_wr_control = value;
+   if (p->devinfo->gen >= 6)
+      brw_inst_set_acc_wr_control(p->devinfo, p->current, value);
 }
 
-void brw_push_insn_state( struct brw_compile *p )
+void brw_push_insn_state( struct brw_codegen *p )
 {
    assert(p->current != &p->stack[BRW_EU_MAX_INSN_STACK-1]);
-   memcpy(p->current+1, p->current, sizeof(struct brw_instruction));
+   memcpy(p->current + 1, p->current, sizeof(brw_inst));
    p->compressed_stack[p->current - p->stack] = p->compressed;
    p->current++;
 }
 
-void brw_pop_insn_state( struct brw_compile *p )
+void brw_pop_insn_state( struct brw_codegen *p )
 {
    assert(p->current != p->stack);
    p->current--;
@@ -177,18 +209,19 @@ void brw_pop_insn_state( struct brw_compile *p )
 /***********************************************************************
  */
 void
-brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
+brw_init_codegen(const struct brw_device_info *devinfo,
+                 struct brw_codegen *p, void *mem_ctx)
 {
    memset(p, 0, sizeof(*p));
 
-   p->brw = brw;
+   p->devinfo = devinfo;
    /*
     * Set the initial instruction store array size to 1024, if found that
     * isn't enough, then it will double the store size at brw_next_insn()
     * until out of memory.
     */
    p->store_size = 1024;
-   p->store = rzalloc_array(mem_ctx, struct brw_instruction, p->store_size);
+   p->store = rzalloc_array(mem_ctx, brw_inst, p->store_size);
    p->nr_insn = 0;
    p->current = p->stack;
    p->compressed = false;
@@ -198,10 +231,10 @@ brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
 
    /* Some defaults?
     */
-   brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
-   brw_set_saturate(p, 0);
-   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
-   brw_set_predicate_control_flag_value(p, 0xff);
+   brw_set_default_exec_size(p, BRW_EXECUTE_8);
+   brw_set_default_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
+   brw_set_default_saturate(p, 0);
+   brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
 
    /* Set up control flow stack */
    p->if_stack_depth = 0;
@@ -213,53 +246,52 @@ brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
    p->loop_stack = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
    p->if_depth_in_loop = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
 
-   brw_init_compaction_tables(brw);
+   brw_init_compaction_tables(devinfo);
 }
 
 
-const unsigned *brw_get_program( struct brw_compile *p,
+const unsigned *brw_get_program( struct brw_codegen *p,
                               unsigned *sz )
 {
-   brw_compact_instructions(p);
-
    *sz = p->next_insn_offset;
    return (const unsigned *)p->store;
 }
 
 void
-brw_dump_compile(struct brw_compile *p, FILE *out, int start, int end)
+brw_disassemble(const struct brw_device_info *devinfo,
+                void *assembly, int start, int end, FILE *out)
 {
-   struct brw_context *brw = p->brw;
-   void *store = p->store;
    bool dump_hex = false;
 
    for (int offset = start; offset < end;) {
-      struct brw_instruction *insn = store + offset;
-      struct brw_instruction uncompacted;
-      printf("0x%08x: ", offset);
-
-      if (insn->header.cmpt_control) {
-        struct brw_compact_instruction *compacted = (void *)insn;
+      brw_inst *insn = assembly + offset;
+      brw_inst uncompacted;
+      bool compacted = brw_inst_cmpt_control(devinfo, insn);
+      if (0)
+         fprintf(out, "0x%08x: ", offset);
+
+      if (compacted) {
+         brw_compact_inst *compacted = (void *)insn;
         if (dump_hex) {
-           printf("0x%08x 0x%08x                       ",
-                  ((uint32_t *)insn)[1],
-                  ((uint32_t *)insn)[0]);
+           fprintf(out, "0x%08x 0x%08x                       ",
+                   ((uint32_t *)insn)[1],
+                   ((uint32_t *)insn)[0]);
         }
 
-        brw_uncompact_instruction(brw, &uncompacted, compacted);
+        brw_uncompact_instruction(devinfo, &uncompacted, compacted);
         insn = &uncompacted;
         offset += 8;
       } else {
         if (dump_hex) {
-           printf("0x%08x 0x%08x 0x%08x 0x%08x ",
-                  ((uint32_t *)insn)[3],
-                  ((uint32_t *)insn)[2],
-                  ((uint32_t *)insn)[1],
-                  ((uint32_t *)insn)[0]);
+           fprintf(out, "0x%08x 0x%08x 0x%08x 0x%08x ",
+                   ((uint32_t *)insn)[3],
+                   ((uint32_t *)insn)[2],
+                   ((uint32_t *)insn)[1],
+                   ((uint32_t *)insn)[0]);
         }
         offset += 16;
       }
 
-      brw_disasm(stdout, insn, p->brw->gen);
+      brw_disassemble_inst(out, devinfo, insn, compacted);
    }
 }